Merge pull request #74 from mapingshuo/fl-mobile

add mobile

python/paddle_fl/mobile/Makefile

+clean:
+	rm -rf *~ *pyc */*~ */*.pyc

python/paddle_fl/mobile/README.md

+
+## 联邦算法模拟器 (fl-mobile simulator)
+
+FL-mobile是一个集移动端算法模拟调研、训练和部署为一体的框架。算法模拟器 (simulator) 是FL-mobile的一部分。
+
+该模拟器的设计目的，是为了模拟实际线上多个移动端设备配合训练的场景。框架的设计思想在服务器上模拟数个端上设备，快速验证算法效果。模拟器的优势为：
+
+- 支持单机和分布式训练
+- 支持常见开源数据集的训练
+- 支持模型中的私有参数和共享参数，私有参数不参与全局更新
+
+## 准备工作
+
+- 安装mpirun
+- python安装grpc
+    ```shell
+    pip install grpcio==1.28.1
+    ```
+- 安装Paddle
+
+## 快速开始
+
+我们以Leaf数据集中的[reddit数据](https://github.com/TalwalkarLab/leaf/tree/master/data/reddit)为例，用LSTM建模，在simulator
+中给出一个单机训练的例子，通过这个例子，您能了解simulator的基础用法。
+
+### 准备数据
+
+```
+wget https://paddle-serving.bj.bcebos.com/temporary_files_for_docker/reddit_subsampled.zip --no-check-certificate
+unzip reddit_subsampled.zip
+```
+在模拟器中，我们假设用户的数据都是天级别的，因此我们将下载的数据重新归置如下
+
+```
+tree lm_data
+lm_data
+|-- 20200101
+|   `-- train_data.json
+|-- 20200102
+|   `-- test_data.json
+`-- vocab.json
+```
+可以看到，我们将训练数据作为20200101的数据，测试数据作为20200102的数据。
+
+### 生成server代码
+
+```
+cd protos
+python run_codegen.py
+cd ..
+```
+
+### 开始训练
+
+```shell
+export PYTHONPATH=$PWD:$PYTHONPATH
+mpirun -np 2 python application.py lm_data
+```
+
+### 训练结果
+
+```shell
+framework.py : INFO  infer results: 0.085723
+```
+
+即：在测试集上的，测试Top1为 8.6% 
+
+## 添加自己的数据集和Trainer
+
+如果您想要训练自己的联邦模型，您需要做四件事：
+
+1. 创建reader，参考`reader/leaf_reddit_reader.py`
+2. 创建trainer，参考`trainer/language_model_trainer.py`
+3. 创建model，即组网，参考`model/language_model.py`
+4. 创建application，参考`application.py`
+
+## 模拟器(simulator) 介绍
+
+框架主要由scheduler和simulator构成；其中scheduler负责统筹规划数据和全局参数；simulator负责做实际的训练和私有参数更新。
+
+- scheduler
+在一次训练流程中，只会有一个global scheduler， 而每个机器上都会有一个scheduler client，负责和global scheduler做参数、数据的通信。
+
+- simulator
+每个机器上都会有1个Simulator，每个Simulator又会有多个shard，shard是用于本机并行训练的。
+
+作为一个分布式框架，FL-mobile simulator 也是包含模型初始化、模型分发、模型训练、模型更新四部分，下面通过一次实际的训练流程，来了解一下各个模型的工作吧：
+
+- Step1 模型初始化
+
+    1. 全局参数初始化：由编号为0的simulator来做模型初始化工作，初始化之后，它会通过UpdateGlobalParams()接口将参数传递给Scheduler；
+   
+    2. 个性化参数初始化
+
+- Step2 模型分发
+
+    1. 全局参数分发：每个simulator开始训练之前，都需要先找SchedulerServer拿全局参数，即通过scheduler_client.get_global_params()获得全局参数；
+
+    2. 个性化参数分发：个性化参数由每个local trainer训练前，向data server获取，获取接口为：get_param_by_uid；
+
+- Step3 模型训练
+
+    模型训练是整个流程的核心，多个trainer并行训练，训练中参数不做同步；每个trainer的训练步数由这个用户的数据量决定。
+
+- Step4 模型更新
+
+    1. 全局参数更新：在所有trainer训练结束后，会做一次同步，并且通过FedAvg算法计算参数梯度；之后上传梯度至scheduler，再拉取新的全局参数，回到第二步；
+    
+    2. 个性化参数更新：个性化参数更新简单一些，每个trainer调用set_param_by_uid就可完成自己的个性化参数更新；

python/paddle_fl/mobile/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

python/paddle_fl/mobile/application.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+'''
+For large scale mobile devices simulation, we need to consider:
+- model structure and input features
+- trainer, i.e. how to design training strategy for each simulation
+- user sampler, i.e. users that need to participate training
+- optimizer, i.e. how to update global weights
+
+Currently, we couple trainer and model together for simplicity
+'''
+from utils import FLSimRoleMaker
+from framework import SimulationFramework
+from trainer import LanguageModelTrainer
+from optimizer import FedAvgOptimizer
+from sampler import UniformSampler, Test1percentSampler
+from datetime import date, timedelta
+import sys
+import time
+
+role_maker = FLSimRoleMaker()
+role_maker.init_env(local_shard_num=30)
+simulator = SimulationFramework(role_maker)
+
+language_model_trainer = LanguageModelTrainer()
+
+language_model_trainer.set_trainer_configs({
+    "epoch": 3,
+    "max_steps_in_epoch": -1,
+    "lr": 0.1,
+    "batch_size": 5,
+})
+
+sampler = UniformSampler()
+sampler.set_sample_num(30)
+sampler.set_min_ins_num(1)
+test_sampler = Test1percentSampler()
+fed_avg_optimizer = FedAvgOptimizer(learning_rate=2.0)
+
+simulator.set_trainer(language_model_trainer)
+simulator.set_sampler(sampler)
+simulator.set_test_sampler(test_sampler)
+simulator.set_fl_optimizer(fed_avg_optimizer)
+
+if simulator.is_scheduler():
+    simulator.run_scheduler_service()
+elif simulator.is_simulator():
+    base_path = sys.argv[1]
+    dates = []
+    start_date = date(2020, 1, 1)
+    end_date = date(2020, 1, 2)
+    delta = timedelta(days=1)
+    while start_date <= end_date:
+        dates.append(start_date.strftime("%Y%m%d"))
+        start_date += delta
+
+    print("dates: {}".format(dates))
+
+    time.sleep(10)
+    simulator.run_simulation(
+        base_path, dates, sim_num_everyday=100, do_test=True, test_skip_day=1)

python/paddle_fl/mobile/clients/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .data_client_impl import DataClient
+from .scheduler_client_impl import SchedulerClient

python/paddle_fl/mobile/clients/data_client_impl.py

+# -*- coding: utf-8 -*- 
+from __future__ import print_function
+import grpc
+import servers.data_server_pb2 as data_server_pb2
+import servers.data_server_pb2_grpc as data_server_pb2_grpc
+from concurrent import futures
+from multiprocessing import Process
+from utils.hdfs_utils import HDFSClient, multi_download
+import time
+import sys
+import os
+import xxhash
+import numpy as np
+from utils.logger import logging
+
+
+class DataClient(object):
+    def __init__(self):
+        self.stub_list = []
+        self.load_data_into_patch = None
+
+    def uid_shard(self, uid):
+        try:
+            uid_hash = xxhash.xxh32(str(uid), seed=101).intdigest()
+        except:
+            return -1
+        shard_idx = uid_hash % len(self.stub_list)
+        return shard_idx
+
+    # should set all params to numpy array with shape and dtype
+    # buggy here
+    def set_param_by_uid(self, uid, param_dict):
+        shard_idx = self.uid_shard(uid)
+        if shard_idx == -1:
+            return -1
+        user_param = data_server_pb2.UserParams()
+        user_param.uid = uid
+        for key in param_dict:
+            param = data_server_pb2.Param()
+            param.name = key
+            np_var = param_dict[param.name]
+            param.shape.extend(np_var.shape)
+            param.weight.extend(np_var.ravel())
+            user_param.user_params.extend([param])
+
+        call_future = self.stub_list[shard_idx].UpdateUserParams.future(
+            user_param)
+        err_code = call_future.result().err_code
+        return err_code
+
+    def get_param_by_uid(self, uid):
+        shard_idx = self.uid_shard(uid)
+        if shard_idx == -1:
+            return -1
+        data = data_server_pb2.Data()
+        data.uid = uid
+        call_future = self.stub_list[shard_idx].GetUserParams.future(data)
+        user_params = call_future.result()
+        param_dict = {}
+        for param in user_params.user_params:
+            param_dict[param.name] = np.array(
+                list(param.weight), dtype=np.float32)
+            param_dict[param.name].shape = list(param.shape)
+        return param_dict
+
+    def clear_user_data(self, date):
+        def clear():
+            for stub in self.stub_list:
+                data = data_server_pb2.Data()
+                data.date = date
+                call_future = stub.ClearUserData.future(data)
+                res = call_future.result()
+
+        p = Process(target=clear, args=())
+        p.start()
+        p.join()
+
+    def get_data_by_uid(self, uid, date):
+        shard_idx = self.uid_shard(uid)
+        if shard_idx == -1:
+            return -1
+        data = data_server_pb2.Data()
+        data.uid = uid
+        data.date = date
+        call_future = self.stub_list[shard_idx].GetUserData.future(data)
+        user_data_list = []
+        for item in call_future.result().line_str:
+            user_data_list.append(item)
+        return user_data_list
+
+    def set_data_server_endpoints(self, endpoints):
+        self.stub_list = []
+        for ep in endpoints:
+            options = [('grpc.max_message_length', 1024 * 1024 * 1024),
+                       ('grpc.max_receive_message_length', 1024 * 1024 * 1024)]
+            channel = grpc.insecure_channel(ep, options=options)
+            stub = data_server_pb2_grpc.DataServerStub(channel)
+            self.stub_list.append(stub)
+
+    def global_shuffle_by_patch(self, data_patch, date, concurrency):
+        shuffle_time = len(data_patch) / concurrency + 1
+        for i in range(shuffle_time):
+            if i * concurrency >= len(data_patch):
+                break
+            pros = []
+            end = min((i + 1) * concurrency, len(data_patch))
+            patch_list = data_patch[i * concurrency:end]
+            width = len(patch_list)
+            for j in range(width):
+                p = Process(
+                    target=self.send_one_patch, args=(patch_list[j], date))
+                pros.append(p)
+            for p in pros:
+                p.start()
+            for p in pros:
+                p.join()
+            logging.info("shuffle round {} done.".format(i))
+
+    def send_one_patch(self, patch, date):
+        for line in patch:
+            group = line.strip().split("\t")
+            if len(group) != 3:
+                continue
+            data = data_server_pb2.Data()
+            data.uid = group[0]
+            data.date = date
+            data.line = line.strip()
+            stub_idx = self.uid_shard(data.uid)
+            if stub_idx == -1:
+                logging.info("send_one_patch continue for uid: %s" % data.uid)
+                continue
+            call_future = self.stub_list[stub_idx].SendData.future(data)
+            u_num = call_future.result()
+
+    def global_shuffle_by_file(self, filelist, concurrency):
+        pass
+
+    def set_load_data_into_patch_func(self, func):
+        self.load_data_into_patch = func
+
+    def get_local_files(self,
+                        base_path,
+                        date,
+                        node_idx,
+                        node_num,
+                        hdfs_configs=None):
+        full_path = "{}/{}".format(base_path, date)
+        if os.path.exists(full_path):
+            file_list = os.listdir(full_path)
+            local_files = ["{}/{}".format(full_path, x) for x in file_list]
+        elif hdfs_configs is not None:
+            local_files = self.download_from_hdfs(hdfs_configs, base_path,
+                                                  date, node_idx, node_num)
+        else:
+            local_files = []
+        return local_files
+
+    def download_from_hdfs(self, hdfs_configs, base_path, date, node_idx,
+                           node_num):
+        # return local filelist
+        hdfs_client = HDFSClient("$HADOOP_HOME", hdfs_configs)
+        multi_download(
+            hdfs_client,
+            "{}/{}".format(base_path, date),
+            date,
+            node_idx,
+            node_num,
+            multi_processes=30)
+        filelist = os.listdir(date)
+        files = ["{}/{}".format(date, fn) for fn in filelist]
+        return files
+
+
+def test_global_shuffle():
+    data_client = DataClient()
+    server_endpoints = ["127.0.0.1:{}".format(50050 + i) for i in range(10)]
+    data_client.set_data_server_endpoints(server_endpoints)
+    date = "0330"
+    file_name = ["data_with_uid/part-01991"]
+    with open(file_name[0]) as fin:
+        for line in fin:
+            group = line.strip().split("\t")
+            uid = group[0]
+            user_data_dict = data_client.get_data_by_uid(uid)
+
+
+def test_set_param():
+    data_client = DataClient()
+    server_endpoints = ["127.0.0.1:{}".format(50050 + i) for i in range(10)]
+    data_client.set_data_server_endpoints(server_endpoints)
+    uid = ["1001", "10001", "100001", "101"]
+    param_dict = {"w0": [1.0, 1.1, 1.2, 1.3], "b0": [1.1, 1.2, 1.3, 1.5]}
+    for cur_i in uid:
+        data_client.set_param_by_uid(cur_i, param_dict)
+
+
+def test_get_param():
+    data_client = DataClient()
+    server_endpoints = ["127.0.0.1:{}".format(50050 + i) for i in range(10)]
+    data_client.set_data_server_endpoints(server_endpoints)
+    uid = ["1001", "10001", "100001", "101"]
+    for cur_i in uid:
+        param_dict = data_client.get_param_by_uid(cur_i)
+        print(param_dict)
+
+
+if __name__ == "__main__":
+    #load_data_global_shuffle()
+    #test_global_shuffle()
+    test_set_param()
+    test_get_param()

python/paddle_fl/mobile/clients/scheduler_client_impl.py

+# -*- coding: utf-8 -*-
+from __future__ import print_function
+import grpc
+import servers.scheduler_server_pb2 as scheduler_server_pb2
+import servers.scheduler_server_pb2_grpc as scheduler_server_pb2_grpc
+import servers.data_server_pb2_grpc as data_server_pb2_grpc
+import servers.data_server_pb2 as data_server_pb2
+import numpy as np
+from concurrent import futures
+from multiprocessing import Process
+
+import time
+import sys
+import os
+
+
+class SchedulerClient(object):
+    def __init__(self):
+        self.stub = None
+        self.stub_list = []
+        self.global_param_info = {}
+
+    def update_user_inst_num(self, date, user_info_dict):
+        user_info = scheduler_server_pb2.UserInstInfo()
+        for key in user_info_dict:
+            single_user_info = scheduler_server_pb2.UserInstNum()
+            single_user_info.uid = key
+            single_user_info.inst_num = user_info_dict[key]
+            user_info.inst_nums.extend([single_user_info])
+        user_info.shard_num = len(self.stub_list)
+        user_info.date = date
+        call_future = self.stub.UpdateUserInstNum.future(user_info)
+        res = call_future.result()
+        return res.err_code
+
+    def set_scheduler_server_endpoints(self, endpoints):
+        options = [('grpc.max_message_length', 1024 * 1024 * 1024),
+                   ('grpc.max_receive_message_length', 1024 * 1024 * 1024)]
+        channel = grpc.insecure_channel(endpoints[0], options=options)
+        self.stub = scheduler_server_pb2_grpc.SchedulerServerStub(channel)
+
+    def set_data_server_endpoints(self, endpoints):
+        self.stub_list = []
+        for ep in endpoints:
+            options = [('grpc.max_message_length', 1024 * 1024 * 1024),
+                       ('grpc.max_receive_message_length', 1024 * 1024 * 1024)]
+            channel = grpc.insecure_channel(ep, options=options)
+            stub = data_server_pb2_grpc.DataServerStub(channel)
+            self.stub_list.append(stub)
+
+    def uniform_sample_user_list(self, date, node_id, sample_num, shard_num,
+                                 node_num, min_ins_num):
+        user_info_dict = {}
+        req = scheduler_server_pb2.Request()
+        req.node_idx = node_id
+        req.sample_num = sample_num
+        req.shard_num = shard_num
+        req.node_num = node_num
+        req.date = date
+        req.min_ins_num = min_ins_num
+        call_future = self.stub.SampleUsersToTrain.future(req)
+        user_info = call_future.result()
+        for user in user_info.inst_nums:
+            user_info_dict[user.uid] = user.inst_num
+        return user_info_dict
+
+    def hash_sample_user_list(self, date, node_id, sample_num, shard_num,
+                              node_num):
+        user_info_dict = {}
+        req = scheduler_server_pb2.Request()
+        req.node_idx = node_id
+        req.sample_num = sample_num
+        req.shard_num = shard_num
+        req.node_num = node_num
+        req.date = date
+        call_future = self.stub.SampleUsersWithHash.future(req)
+        user_info = call_future.result()
+        for user in user_info.inst_nums:
+            user_info_dict[user.uid] = user.inst_num
+        return user_info_dict
+
+    def sample_test_user_list(self, date, node_id, shard_num, node_num):
+        user_info_dict = {}
+        req = scheduler_server_pb2.Request()
+        req.node_idx = node_id
+        req.shard_num = shard_num
+        req.node_num = node_num
+        req.date = date
+        call_future = self.stub.SampleUsersToTest.future(req)
+        user_info = call_future.result()
+        for user in user_info.inst_nums:
+            user_info_dict[user.uid] = user.inst_num
+        return user_info_dict
+
+    def fixed_sample_user_list(self, date, node_id, sample_num, shard_num,
+                               node_num):
+        user_info_dict = {}
+        req = scheduler_server_pb2.Request()
+        req.node_idx = node_id
+        req.sample_num = sample_num
+        req.shard_num = shard_num
+        req.node_num = node_num
+        req.date = date
+        call_future = self.stub.FixedUsersToTrain.future(req)
+        user_info = call_future.result()
+        for user in user_info.inst_nums:
+            user_info_dict[user.uid] = user.inst_num
+        return user_info_dict
+
+    def get_global_params(self):
+        req = scheduler_server_pb2.Request()
+        req.node_idx = 0
+        req.sample_num = 0
+        req.shard_num = 0
+        req.node_num = 0
+        call_future = self.stub.GetGlobalParams.future(req)
+        global_p = call_future.result()
+        result_dict = {}
+        for param in global_p.global_params:
+            result_dict[param.name] = np.array(
+                list(param.weight), dtype=np.float32)
+            result_dict[param.name].shape = param.shape
+        return result_dict
+
+    def update_global_params(self, global_params):
+        global_p = scheduler_server_pb2.GlobalParams()
+        for key in global_params:
+            param = scheduler_server_pb2.Param()
+            param.name = key
+            var, shape = global_params[key], global_params[key].shape
+            self.global_param_info[key] = shape
+            param.weight.extend(var.ravel())
+            param.shape.extend(shape)
+            global_p.global_params.extend([param])
+        call_future = self.stub.UpdateGlobalParams.future(global_p)
+        res = call_future.result()
+        return res.err_code
+
+    def fedavg_update(self, global_param_delta_dict):
+        global_p = scheduler_server_pb2.GlobalParams()
+        for key in global_param_delta_dict:
+            param = scheduler_server_pb2.Param()
+            param.name = key
+            parameter_delta, shape = global_param_delta_dict[
+                param.name], global_param_delta_dict[param.name].shape
+            param.weight.extend(parameter_delta.ravel())
+            param.shape.extend(shape)
+            global_p.global_params.extend([param])
+        call_future = self.stub.FedAvgUpdate.future(global_p)
+        res = call_future.result()
+        return res.err_code
+
+    def stop_scheduler_server(self):
+        empty_input = scheduler_server_pb2.SchedulerServerEmptyInput()
+        call_future = self.stub.Exit.future(empty_input)
+        res = call_future.result()
+
+
+def test_uniform_sample_user_list():
+    client = SchedulerClient()
+    client.set_scheduler_server_endpoints(["127.0.0.1:60001"])
+    # buggy
+    #user_list = client.get_user_list()
+    global_user_list = []
+    for i in range(10000):
+        global_user_list.append((str(i), 100))
+    client.update_user_inst_num(global_user_list)
+    user_info = client.uniform_sample_user_list(0)
+    user_list = [("101", 100), ("102", 100), ("103", 10000)]
+    global_param = {"w0": [1.0, 1.0, 1.0], "w1": [2.0, 2.0, 2.0]}
+    client.update_global_params(global_param)
+    fetched_params = client.get_global_params()
+
+
+def test_get_global_params():
+    client = SchedulerClient()
+    client.set_scheduler_server_endpoints(["127.0.0.1:60001"])
+    global_param = {"w0": [1.0, 1.0, 1.0], "w1": [2.0, 2.0, 2.0]}
+    client.update_global_params(global_param)
+    fetched_params = client.get_global_params()
+    print(fetched_params)
+
+
+def test_update_global_params():
+    client = SchedulerClient()
+    client.set_scheduler_server_endpoints(["127.0.0.1:60001"])
+    global_param = {"w0": [1.0, 1.0, 1.0], "w1": [3.0, 3.0, 3.0]}
+    client.update_global_params(global_param)
+    fetched_params = client.get_global_params()
+    print(fetched_params)
+
+
+if __name__ == "__main__":
+    test_uniform_sample_user_list()
+    test_update_global_params()
+    test_get_global_params()

python/paddle_fl/mobile/framework.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from utils.role_maker import FLSimRoleMaker
+from clients import DataClient
+from clients import SchedulerClient
+from servers import DataServer
+from servers import SchedulerServer
+from multiprocessing import Process, Pool, Manager, Pipe, Lock
+from utils.logger import logging
+import pickle
+import time
+import numpy as np
+import sys
+import os
+
+
+class SimulationFramework(object):
+    def __init__(self, role_maker):
+        self.data_client = None
+        self.scheduler_client = None
+        self.role_maker = role_maker
+        # we suppose currently we train homogeneous model
+        self.trainer = None
+        # for sampling users
+        self.sampler = None
+        # for update global weights
+        self.fl_optimizer = None
+        # for samping users to test
+        self.test_sampler = None
+        self.profile_file = open("profile", "w")
+        self.do_profile = True
+        # for data downloading
+        self.hdfs_configs = None
+
+    def set_hdfs_configs(self, configs):
+        self.hdfs_configs = configs
+
+    def set_trainer(self, trainer):
+        self.trainer = trainer
+
+    def set_sampler(self, sampler):
+        self.sampler = sampler
+
+    def set_test_sampler(self, sampler):
+        self.test_sampler = sampler
+
+    def set_fl_optimizer(self, optimizer):
+        self.fl_optimizer = optimizer
+
+    def is_scheduler(self):
+        return self.role_maker.is_global_scheduler()
+
+    def is_simulator(self):
+        return self.role_maker.is_simulator()
+
+    def run_scheduler_service(self):
+        if self.role_maker.is_global_scheduler():
+            self._run_global_scheduler()
+
+    def _barrier_simulators(self):
+        self.role_maker.barrier_simulator()
+
+    def _start_data_server(self, endpoint):
+        data_server = DataServer()
+        port = endpoint.split(":")[1]
+        data_server.start(endpoint=port)
+
+    def _run_global_scheduler(self):
+        scheduler_server = SchedulerServer()
+        endpoint = self.role_maker.get_global_scheduler_endpoint()
+        port = endpoint.split(":")[1]
+        scheduler_server.start(endpoint=port)
+
+    def _get_data_services(self):
+        data_server_endpoints = \
+          self.role_maker.get_local_data_server_endpoint()
+        data_server_pros = []
+        for i, ep in enumerate(data_server_endpoints):
+            p = Process(target=self._start_data_server, args=(ep, ))
+            data_server_pros.append(p)
+
+        return data_server_pros
+
+    def _profile(self, func, *args, **kwargs):
+        if self.do_profile:
+            start = time.time()
+            res = func(*args, **kwargs)
+            end = time.time()
+            self.profile_file.write("%s\t\t%f s\n" %
+                                    (func.__name__, end - start))
+            return res
+        else:
+            return func(*args, **kwargs)
+
+    def _run_sim(self, date, sim_num_everyday=1):
+        sim_idx = self.role_maker.simulator_idx()
+        sim_num = self.role_maker.simulator_num()
+        sim_all_trainer_run_time = 0
+        sim_read_praram_and_optimize = 0
+        for sim in range(sim_num_everyday):
+            logging.info("sim id: %d" % sim)
+            # sampler algorithm
+            user_info_dict = self._profile(
+                self.sampler.sample_user_list, self.scheduler_client, date,
+                sim_idx, len(self.data_client.stub_list), sim_num)
+
+            if self.do_profile:
+                print("sim_idx: ", sim_idx)
+                print("shard num: ", len(self.data_client.stub_list))
+                print("sim_num: ", sim_num)
+                print("user_info_dict: ", user_info_dict)
+
+            global_param_dict = self._profile(
+                self.scheduler_client.get_global_params)
+            processes = []
+            os.system("rm -rf _global_param")
+            os.system("mkdir _global_param")
+            start = time.time()
+            for idx, user in enumerate(user_info_dict):
+                arg_dict = {
+                    "uid": str(user),
+                    "date": date,
+                    "data_endpoints":
+                    self.role_maker.get_data_server_endpoints(),
+                    "global_params": global_param_dict,
+                    "user_param_names": self.trainer.get_user_param_names(),
+                    "global_param_names":
+                    self.trainer.get_global_param_names(),
+                    "write_global_param_file":
+                    "_global_param/process_%d" % idx,
+                }
+                p = Process(
+                    target=self.trainer.train_one_user_func,
+                    args=(arg_dict, self.trainer.trainer_config))
+                p.start()
+                processes.append(p)
+            if self.do_profile:
+                logging.info("wait processes to close")
+            for i, p in enumerate(processes):
+                processes[i].join()
+            end = time.time()
+            sim_all_trainer_run_time += (end - start)
+
+            start = time.time()
+            train_result = []
+            new_global_param_by_user = {}
+            training_sample_by_user = {}
+            for i, p in enumerate(processes):
+                param_dir = "_global_param/process_%d/" % i
+                with open(param_dir + "/_info", "r") as f:
+                    user, train_sample_num = pickle.load(f)
+                param_dict = {}
+                for f_name in os.listdir(os.path.join(param_dir, "params")):
+                    f_path = os.path.join(param_dir, "params", f_name)
+                    if os.path.isdir(f_path):  # layer
+                        for layer_param in os.listdir(f_path):
+                            layer_param_path = os.path.join(f_path,
+                                                            layer_param)
+                            with open(layer_param_path) as f:
+                                param_dict["{}/{}".format(
+                                    f_name, layer_param)] = np.load(f)
+                    else:
+                        with open(f_path) as f:
+                            param_dict[f_name] = np.load(f)
+                new_global_param_by_user[user] = param_dict
+                training_sample_by_user[user] = train_sample_num
+
+            self.fl_optimizer.update(training_sample_by_user,
+                                     new_global_param_by_user,
+                                     global_param_dict, self.scheduler_client)
+            end = time.time()
+            sim_read_praram_and_optimize += (end - start)
+        if self.do_profile:
+            self.profile_file.write("sim_all_trainer_run_time\t\t%f s\n" %
+                                    sim_all_trainer_run_time)
+            self.profile_file.write("sim_read_praram_and_optimize\t\t%f s\n" %
+                                    sim_read_praram_and_optimize)
+
+        logging.info("training done for date %s." % date)
+
+    def _test(self, date):
+        if self.trainer.infer_one_user_func is None:
+            pass
+        logging.info("doing test...")
+        if self.test_sampler is None:
+            logging.error("self.test_sampler should not be None when testing")
+
+        sim_idx = self.role_maker.simulator_idx()
+        sim_num = self.role_maker.simulator_num()
+        user_info_dict = self.test_sampler.sample_user_list(
+            self.scheduler_client,
+            date,
+            sim_idx,
+            len(self.data_client.stub_list),
+            sim_num, )
+        if self.do_profile:
+            print("test user info_dict: ", user_info_dict)
+        global_param_dict = self.scheduler_client.get_global_params()
+
+        def divide_chunks(l, n):
+            for i in range(0, len(l), n):
+                yield l[i:i + n]
+
+        # at most 50 process for testing
+        chunk_size = 50
+        # at most 100 uid for testing
+        max_test_uids = 100
+        uid_chunks = divide_chunks(user_info_dict.keys(), chunk_size)
+        os.system("rm -rf _test_result")
+        os.system("mkdir _test_result")
+
+        tested_uids = 0
+        for uids in uid_chunks:
+            if tested_uids >= max_test_uids:
+                break
+            processes = []
+            for user in uids:
+                arg_dict = {
+                    "uid": str(user),
+                    "date": date,
+                    "data_endpoints":
+                    self.role_maker.get_data_server_endpoints(),
+                    "global_params": global_param_dict,
+                    "user_param_names": self.trainer.get_user_param_names(),
+                    "global_param_names":
+                    self.trainer.get_global_param_names(),
+                    "infer_result_dir": "_test_result/uid-%s" % user,
+                }
+                p = Process(
+                    target=self.trainer.infer_one_user_func,
+                    args=(arg_dict, self.trainer.trainer_config))
+                p.start()
+                processes.append(p)
+            if self.do_profile:
+                logging.info("wait test processes to close")
+            for i, p in enumerate(processes):
+                processes[i].join()
+            tested_uids += chunk_size
+
+        infer_results = []
+        # only support one test metric now
+        for uid in os.listdir("_test_result"):
+            with open("_test_result/" + uid + "/res", 'r') as f:
+                sample_cout, metric = f.readlines()[0].strip('\n').split('\t')
+                infer_results.append((int(sample_cout), float(metric)))
+        if sum([x[0] for x in infer_results]) == 0:
+            logging.info("infer results: 0.0")
+        else:
+            count = sum([x[0] for x in infer_results])
+            metric = sum([x[0] * x[1] for x in infer_results]) / count
+            logging.info("infer results: %f" % metric)
+
+    def _save_and_upload(self, date, fs_upload_path):
+        if self.trainer.save_and_upload_func is None:
+            return
+        if fs_upload_path is None:
+            return
+        dfs_upload_path = fs_upload_path + date + "_" + str(
+            self.role_maker.simulator_idx())
+        global_param_dict = self.scheduler_client.get_global_params()
+        arg_dict = {
+            "date": date,
+            "global_params": global_param_dict,
+            "user_param_names": self.trainer.get_user_param_names(),
+            "global_param_names": self.trainer.get_global_param_names(),
+        }
+        self.trainer.save_and_upload_func(
+            arg_dict, self.trainer.trainer_config, dfs_upload_path)
+
+    def run_simulation(self,
+                       base_path,
+                       dates,
+                       fs_upload_path=None,
+                       sim_num_everyday=1,
+                       do_test=False,
+                       test_skip_day=6):
+        if not self.role_maker.is_simulator():
+            pass
+        data_services = self._get_data_services()
+        for service in data_services:
+            service.start()
+        self._barrier_simulators()
+        self.data_client = DataClient()
+        self.data_client.set_load_data_into_patch_func(
+            self.trainer.get_load_data_into_patch_func())
+        self.data_client.set_data_server_endpoints(
+            self.role_maker.get_data_server_endpoints())
+        self.scheduler_client = SchedulerClient()
+        self.scheduler_client.set_data_server_endpoints(
+            self.role_maker.get_data_server_endpoints())
+        self.scheduler_client.set_scheduler_server_endpoints(
+            [self.role_maker.get_global_scheduler_endpoint()])
+        logging.info("trainer config: ", self.trainer.trainer_config)
+        self.trainer.prepare(do_test=do_test)
+
+        if self.role_maker.simulator_idx() == 0:
+            self.trainer.init_global_model(self.scheduler_client)
+        self._barrier_simulators()
+
+        for date_idx, date in enumerate(dates):
+            if date_idx > 0:
+                self.do_profile = False
+                self.profile_file.close()
+            logging.info("reading data for date: %s" % date)
+            local_files = self._profile(
+                self.data_client.get_local_files,
+                base_path,
+                date,
+                self.role_maker.simulator_idx(),
+                self.role_maker.simulator_num(),
+                hdfs_configs=self.hdfs_configs)
+
+            logging.info("loading data into patch for date: %s" % date)
+            data_patch, local_user_dict = self._profile(
+                self.data_client.load_data_into_patch, local_files, 10000)
+            logging.info("shuffling data for date: %s" % date)
+            self._profile(self.data_client.global_shuffle_by_patch, data_patch,
+                          date, 30)
+
+            logging.info("updating user inst num for date: %s" % date)
+            self._profile(self.scheduler_client.update_user_inst_num, date,
+                          local_user_dict)
+            self.role_maker.barrier_simulator()
+
+            if do_test and date_idx != 0 and date_idx % test_skip_day == 0:
+                self._barrier_simulators()
+                self._profile(self._test, date)
+                self._barrier_simulators()
+                self._profile(self._save_and_upload, date, fs_upload_path)
+
+            self._run_sim(date, sim_num_everyday=sim_num_everyday)
+            self.role_maker.barrier_simulator()
+            logging.info("clear user data for date: %s" % date)
+            self.data_client.clear_user_data(date)
+
+        self._barrier_simulators()
+        logging.info("training done all date.")
+        logging.info("stoping scheduler")
+        self.scheduler_client.stop_scheduler_server()
+        for pro in data_services:
+            pro.terminate()
+        logging.info("after terminate for all server.")

python/paddle_fl/mobile/get_data.sh

+# reddit_subsampled
+wget https://paddle-serving.bj.bcebos.com/temporary_files_for_docker/reddit_subsampled.zip --no-check-certificate
+unzip reddit_subsampled.zip

python/paddle_fl/mobile/layer/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .layer_base import Embedding
+from .layer_base import SequenceConv
+from .layer_base import Concat
+from .layer_base import Pooling
+from .layer_base import FC
+from .layer_base import CrossEntropySum
+from .layer_base import ACC
+from .layer_base import AUC
+from .data_base import TextData

python/paddle_fl/mobile/layer/data_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle.fluid as fluid
+
+
+class DataBase(object):
+    def __init__(self):
+        pass
+
+
+class TextData(object):
+    def __init__(self):
+        pass
+
+    def create(self, name=None):
+        assert name is not None
+        return fluid.layers.data(
+            name=name, shape=[1], dtype='int64', lod_level=1)

python/paddle_fl/mobile/layer/layer_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle.fluid as fluid
+
+
+class LayerBase(object):
+    def __init__(self):
+        pass
+
+
+class Embedding(LayerBase):
+    def __init__(self):
+        pass
+
+    def attr(self, shapes, name=None):
+        self.vocab_size = shapes[0]
+        self.emb_dim = shapes[1]
+        self.name = name
+
+    def forward(self, i):
+        is_sparse = True
+        if self.name is not None:
+            param_attr = fluid.ParamAttr(name=self.name)
+        else:
+            param_attr = None
+
+        results = []
+        emb = fluid.layers.embedding(
+            input=i,
+            is_sparse=is_sparse,
+            size=[self.vocab_size, self.emb_dim],
+            param_attr=param_attr,
+            padding_idx=0)
+        return emb
+
+
+class SequenceConv(LayerBase):
+    def __init__(self):
+        pass
+
+    def attr(self, name=None):
+        self.num_filters = 64
+        self.win_size = 3
+        self.name = name
+
+    def forward(self, i):
+        if self.name is not None:
+            param_attr = fluid.ParamAttr(name=self.name)
+        else:
+            param_attr = None
+        conv = fluid.nets.sequence_conv_pool(
+            input=i,
+            num_filters=self.num_filters,
+            filter_size=self.win_size,
+            param_attr=param_attr,
+            act="tanh",
+            pool_type="max")
+        return conv
+
+
+class Concat(LayerBase):
+    def forward(self, inputs):
+        concat = fluid.layers.concat(inputs, axis=1)
+        return concat
+
+
+class Pooling(LayerBase):
+    def __init__(self):
+        self.pool_type = 'sum'
+
+    def attr(self, pool_type='sum'):
+        self.pool_type = pool_type
+
+    def forward(self, i):
+        pool = fluid.layers.sequence_pool(input=i, pool_type='sum')
+        return pool
+
+
+class FC(LayerBase):
+    def __init__(self):
+        return
+
+    def attr(self, shapes, act='relu', name=None):
+        self.name = name
+        self.size = shapes[0]
+        self.act = act
+
+    def forward(self, i):
+        if self.name is not None:
+            param_attr = fluid.ParamAttr(name=self.name)
+        else:
+            param_attr = None
+        fc = fluid.layers.fc(input=i,
+                             size=self.size,
+                             act=self.act,
+                             param_attr=param_attr)
+        return fc
+
+
+class CrossEntropySum(LayerBase):
+    def __init__(self):
+        pass
+
+    def forward(self, prediction, label):
+        cost = fluid.layers.cross_entropy(input=prediction, label=label)
+        sum_cost = fluid.layers.reduce_sum(cost)
+        return sum_cost
+
+
+class ACC(LayerBase):
+    def __init__(self):
+        pass
+
+    def forward(self, prediction, label):
+        return fluid.layers.accuracy(input=prediction, label=label)
+
+
+class AUC(LayerBase):
+    def forward(self, prediction, label):
+        auc, batch_auc_var, auc_states = \
+                    fluid.layers.auc(input=prediction, label=label,
+                                     num_thresholds=2 ** 12,
+                                     slide_steps=20)
+        return auc, batch_auc_var

python/paddle_fl/mobile/model/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .language_model import LanguageModel

python/paddle_fl/mobile/model/language_model.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from model import *
+from layer import *
+from .model_base import ModelBase
+import paddle.fluid as fluid
+
+import paddle.fluid.layers as layers
+import paddle.fluid as fluid
+from paddle.fluid.layers.control_flow import StaticRNN as PaddingRNN
+import numpy as np
+from paddle.fluid import ParamAttr
+from paddle.fluid.contrib.layers import basic_lstm
+
+
+class LanguageModel(ModelBase):
+    def __init__(self):
+        # model args
+        self.hidden_size_ = 200
+        self.vocab_size_ = 10000
+        self.num_layers_ = 2
+        self.num_steps_ = 10  # fix
+        self.init_scale_ = 0.1
+        self.dropout_ = 0.0
+        self.rnn_model_ = 'basic_lstm'
+        self.pad_symbol_ = 0
+        self.unk_symbol_ = 1
+
+        # results
+        self.correct_ = None
+        self.prediction_ = None
+        self.loss_ = None
+
+        # private vars
+        self.user_params_ = []
+        self.program_ = None
+        self.startup_program_ = None
+        self.input_name_list_ = None
+        self.target_var_names_ = []
+
+    def get_model_input_names(self):
+        return self.input_name_list_
+
+    def get_model_loss(self):
+        return self.loss_
+
+    def get_model_loss_name(self):
+        return self.loss_.name
+
+    def get_model_metrics(self):
+        return {"correct": self.correct_.name}
+
+    def get_target_names(self):
+        return self.target_var_names_
+
+    def build_model(self, model_configs):
+        hidden_size = self.hidden_size_
+        init_scale = self.init_scale_
+        dropout = self.dropout_
+        num_layers = self.num_layers_
+        num_steps = self.num_steps_
+        pad_symbol = self.pad_symbol_
+        unk_symbol = self.unk_symbol_
+        vocab_size = self.vocab_size_
+        rnn_model = self.rnn_model_
+        x = fluid.data(name="x", shape=[None, num_steps], dtype='int64')
+        y = fluid.data(name="y", shape=[None, num_steps], dtype='int64')
+        x = layers.reshape(x, shape=[-1, num_steps, 1])
+        y = layers.reshape(y, shape=[-1, 1])
+        self.input_name_list_ = ['x', 'y']
+
+        init_hidden = layers.fill_constant_batch_size_like(
+            input=x,
+            shape=[-1, num_layers, hidden_size],
+            value=0,
+            dtype="float32")
+        init_cell = layers.fill_constant_batch_size_like(
+            input=x,
+            shape=[-1, num_layers, hidden_size],
+            value=0,
+            dtype="float32")
+
+        init_hidden = layers.transpose(init_hidden, perm=[1, 0, 2])
+        init_cell = layers.transpose(init_cell, perm=[1, 0, 2])
+
+        init_hidden_reshape = layers.reshape(
+            init_hidden, shape=[num_layers, -1, hidden_size])
+        init_cell_reshape = layers.reshape(
+            init_cell, shape=[num_layers, -1, hidden_size])
+
+        x_emb = layers.embedding(
+            input=x,
+            size=[vocab_size, hidden_size],
+            dtype='float32',
+            is_sparse=False,
+            param_attr=fluid.ParamAttr(
+                name='embedding_para',
+                initializer=fluid.initializer.UniformInitializer(
+                    low=-init_scale, high=init_scale)))
+
+        x_emb = layers.reshape(
+            x_emb, shape=[-1, num_steps, hidden_size], inplace=True)
+        if dropout != None and dropout > 0.0:
+            x_emb = layers.dropout(
+                x_emb,
+                dropout_prob=dropout,
+                dropout_implementation='upscale_in_train')
+
+        if rnn_model == "padding":
+            rnn_out, last_hidden, last_cell = self._padding_rnn(
+                x_emb,
+                len=num_steps,
+                init_hidden=init_hidden_reshape,
+                init_cell=init_cell_reshape)
+        elif rnn_model == "static":
+            rnn_out, last_hidden, last_cell = self._encoder_static(
+                x_emb,
+                len=num_steps,
+                init_hidden=init_hidden_reshape,
+                init_cell=init_cell_reshape)
+        elif rnn_model == "cudnn":
+            x_emb = layers.transpose(x_emb, perm=[1, 0, 2])
+            rnn_out, last_hidden, last_cell = layers.lstm(
+                x_emb,
+                init_hidden_reshape,
+                init_cell_reshape,
+                num_steps,
+                hidden_size,
+                num_layers,
+                is_bidirec=False,
+                default_initializer=fluid.initializer.UniformInitializer(
+                    low=-init_scale, high=init_scale))
+            rnn_out = layers.transpose(rnn_out, perm=[1, 0, 2])
+        elif rnn_model == "basic_lstm":
+            rnn_out, last_hidden, last_cell = basic_lstm(
+                x_emb,
+                init_hidden,
+                init_cell,
+                hidden_size,
+                num_layers=num_layers,
+                batch_first=True,
+                dropout_prob=dropout,
+                param_attr=ParamAttr(
+                    initializer=fluid.initializer.UniformInitializer(
+                        low=-init_scale, high=init_scale)),
+                bias_attr=ParamAttr(
+                    initializer=fluid.initializer.Constant(0.0)),
+                forget_bias=0.0)
+        else:
+            raise Exception("type not support")
+
+        rnn_out = layers.reshape(
+            rnn_out, shape=[-1, num_steps, hidden_size], inplace=True)
+
+        softmax_weight = layers.create_parameter(
+            [hidden_size, vocab_size],
+            dtype="float32",
+            name="softmax_weight",
+            default_initializer=fluid.initializer.UniformInitializer(
+                low=-init_scale, high=init_scale))
+        softmax_bias = layers.create_parameter(
+            [vocab_size],
+            dtype="float32",
+            name='softmax_bias',
+            default_initializer=fluid.initializer.UniformInitializer(
+                low=-init_scale, high=init_scale))
+
+        projection = layers.matmul(rnn_out, softmax_weight)
+        projection = layers.elementwise_add(projection, softmax_bias)
+        projection = layers.reshape(
+            projection, shape=[-1, vocab_size], inplace=True)
+
+        # correct predictions
+        labels_reshaped = fluid.layers.reshape(y, [-1])
+        pred = fluid.layers.cast(
+            fluid.layers.argmax(projection, 1), dtype="int64")
+        correct_pred = fluid.layers.cast(
+            fluid.layers.equal(pred, labels_reshaped), dtype="int64")
+        self.prediction_ = pred
+        self.target_var_names_.append(pred)
+
+        # predicting unknown is always considered wrong
+        unk_tensor = fluid.layers.fill_constant(
+            fluid.layers.shape(labels_reshaped),
+            value=unk_symbol,
+            dtype='int64')
+        pred_unk = fluid.layers.cast(
+            fluid.layers.equal(pred, unk_tensor), dtype="int64")
+        correct_unk = fluid.layers.elementwise_mul(pred_unk, correct_pred)
+
+        # predicting padding is always considered wrong
+        pad_tensor = fluid.layers.fill_constant(
+            fluid.layers.shape(labels_reshaped), value=0, dtype='int64')
+        pred_pad = fluid.layers.cast(
+            fluid.layers.equal(pred, pad_tensor), dtype="int64")
+        correct_pad = fluid.layers.elementwise_mul(pred_pad, correct_pred)
+
+        # acc
+        correct_count = fluid.layers.reduce_sum(correct_pred) \
+                - fluid.layers.reduce_sum(correct_unk) \
+                - fluid.layers.reduce_sum(correct_pad)
+        self.correct_ = correct_count
+        self.target_var_names_.append(correct_count)
+
+        loss = layers.softmax_with_cross_entropy(
+            logits=projection, label=y, soft_label=False)
+
+        loss = layers.reshape(loss, shape=[-1, num_steps], inplace=True)
+        loss = layers.reduce_mean(loss, dim=[0])
+        loss = layers.reduce_sum(loss)
+        self.loss_ = loss
+        self.target_var_names_.append(loss)
+
+        loss.persistable = True
+
+        # This will feed last_hidden, last_cell to init_hidden, init_cell, which
+        # can be used directly in next batch. This can avoid the fetching of
+        # last_hidden and last_cell and feeding of init_hidden and init_cell in
+        # each training step.
+        #last_hidden = layers.transpose(last_hidden, perm=[1, 0, 2])
+        #last_cell = layers.transpose(last_cell, perm=[1, 0, 2])
+        #self.input_name_list_ = ['x', 'y', 'init_hidden', 'init_cell']
+
+        self.program_ = fluid.default_main_program()
+        self.startup_program_ = fluid.default_startup_program()
+
+    def _padding_rnn(input_embedding, len=3, init_hidden=None, init_cell=None):
+        weight_1_arr = []
+        weight_2_arr = []
+        bias_arr = []
+        hidden_array = []
+        cell_array = []
+        mask_array = []
+        hidden_size = self.hidden_size_
+        init_scale = self.init_scale_
+        dropout = slef.dropout_
+        num_layers = self.num_layers_
+        num_steps = self._num_steps_
+        for i in range(num_layers):
+            weight_1 = layers.create_parameter(
+                [hidden_size * 2, hidden_size * 4],
+                dtype="float32",
+                name="fc_weight1_" + str(i),
+                default_initializer=fluid.initializer.UniformInitializer(
+                    low=-init_scale, high=init_scale))
+            weight_1_arr.append(weight_1)
+            bias_1 = layers.create_parameter(
+                [hidden_size * 4],
+                dtype="float32",
+                name="fc_bias1_" + str(i),
+                default_initializer=fluid.initializer.Constant(0.0))
+            bias_arr.append(bias_1)
+
+            pre_hidden = layers.slice(
+                init_hidden, axes=[0], starts=[i], ends=[i + 1])
+            pre_cell = layers.slice(
+                init_cell, axes=[0], starts=[i], ends=[i + 1])
+            pre_hidden = layers.reshape(pre_hidden, shape=[-1, hidden_size])
+            pre_cell = layers.reshape(pre_cell, shape=[-1, hidden_size])
+            hidden_array.append(pre_hidden)
+            cell_array.append(pre_cell)
+
+        input_embedding = layers.transpose(input_embedding, perm=[1, 0, 2])
+        rnn = PaddingRNN()
+
+        with rnn.step():
+            input = rnn.step_input(input_embedding)
+            for k in range(num_layers):
+                pre_hidden = rnn.memory(init=hidden_array[k])
+                pre_cell = rnn.memory(init=cell_array[k])
+                weight_1 = weight_1_arr[k]
+                bias = bias_arr[k]
+
+                nn = layers.concat([input, pre_hidden], 1)
+                gate_input = layers.matmul(x=nn, y=weight_1)
+
+                gate_input = layers.elementwise_add(gate_input, bias)
+                i = layers.slice(
+                    gate_input, axes=[1], starts=[0], ends=[hidden_size])
+                j = layers.slice(
+                    gate_input,
+                    axes=[1],
+                    starts=[hidden_size],
+                    ends=[hidden_size * 2])
+                f = layers.slice(
+                    gate_input,
+                    axes=[1],
+                    starts=[hidden_size * 2],
+                    ends=[hidden_size * 3])
+                o = layers.slice(
+                    gate_input,
+                    axes=[1],
+                    starts=[hidden_size * 3],
+                    ends=[hidden_size * 4])
+
+                c = pre_cell * layers.sigmoid(f) + layers.sigmoid(
+                    i) * layers.tanh(j)
+                m = layers.tanh(c) * layers.sigmoid(o)
+
+                rnn.update_memory(pre_hidden, m)
+                rnn.update_memory(pre_cell, c)
+
+                rnn.step_output(m)
+                rnn.step_output(c)
+
+                input = m
+
+                if dropout != None and dropout > 0.0:
+                    input = layers.dropout(
+                        input,
+                        dropout_prob=dropout,
+                        dropout_implementation='upscale_in_train')
+
+            rnn.step_output(input)
+        rnnout = rnn()
+
+        last_hidden_array = []
+        last_cell_array = []
+        real_res = rnnout[-1]
+        for i in range(num_layers):
+            m = rnnout[i * 2]
+            c = rnnout[i * 2 + 1]
+            m.stop_gradient = True
+            c.stop_gradient = True
+            last_h = layers.slice(
+                m, axes=[0], starts=[num_steps - 1], ends=[num_steps])
+            last_hidden_array.append(last_h)
+            last_c = layers.slice(
+                c, axes=[0], starts=[num_steps - 1], ends=[num_steps])
+            last_cell_array.append(last_c)
+        real_res = layers.transpose(x=real_res, perm=[1, 0, 2])
+        last_hidden = layers.concat(last_hidden_array, 0)
+        last_cell = layers.concat(last_cell_array, 0)
+
+        return real_res, last_hidden, last_cell
+
+    def _encoder_static(input_embedding,
+                        len=3,
+                        init_hidden=None,
+                        init_cell=None):
+        weight_1_arr = []
+        weight_2_arr = []
+        bias_arr = []
+        hidden_array = []
+        cell_array = []
+        mask_array = []
+        hidden_size = self.hidden_size_
+        init_scale = self.init_scale_
+        dropout = slef.dropout_
+        num_layers = self.num_layers_
+        for i in range(num_layers):
+            weight_1 = layers.create_parameter(
+                [hidden_size * 2, hidden_size * 4],
+                dtype="float32",
+                name="fc_weight1_" + str(i),
+                default_initializer=fluid.initializer.UniformInitializer(
+                    low=-init_scale, high=init_scale))
+            weight_1_arr.append(weight_1)
+            bias_1 = layers.create_parameter(
+                [hidden_size * 4],
+                dtype="float32",
+                name="fc_bias1_" + str(i),
+                default_initializer=fluid.initializer.Constant(0.0))
+            bias_arr.append(bias_1)
+
+            pre_hidden = layers.slice(
+                init_hidden, axes=[0], starts=[i], ends=[i + 1])
+            pre_cell = layers.slice(
+                init_cell, axes=[0], starts=[i], ends=[i + 1])
+            pre_hidden = layers.reshape(
+                pre_hidden, shape=[-1, hidden_size], inplace=True)
+            pre_cell = layers.reshape(
+                pre_cell, shape=[-1, hidden_size], inplace=True)
+            hidden_array.append(pre_hidden)
+            cell_array.append(pre_cell)
+
+        res = []
+        sliced_inputs = layers.split(
+            input_embedding, num_or_sections=len, dim=1)
+
+        for index in range(len):
+            input = sliced_inputs[index]
+            input = layers.reshape(
+                input, shape=[-1, hidden_size], inplace=True)
+            for k in range(num_layers):
+                pre_hidden = hidden_array[k]
+                pre_cell = cell_array[k]
+                weight_1 = weight_1_arr[k]
+                bias = bias_arr[k]
+                nn = layers.concat([input, pre_hidden], 1)
+                gate_input = layers.matmul(x=nn, y=weight_1)
+
+                gate_input = layers.elementwise_add(gate_input, bias)
+                i, j, f, o = layers.split(
+                    gate_input, num_or_sections=4, dim=-1)
+
+                c = pre_cell * layers.sigmoid(f) + layers.sigmoid(
+                    i) * layers.tanh(j)
+                m = layers.tanh(c) * layers.sigmoid(o)
+
+                hidden_array[k] = m
+                cell_array[k] = c
+                input = m
+
+                if dropout != None and dropout > 0.0:
+                    input = layers.dropout(
+                        input,
+                        dropout_prob=dropout,
+                        dropout_implementation='upscale_in_train')
+
+            res.append(input)
+
+        last_hidden = layers.concat(hidden_array, 1)
+        last_hidden = layers.reshape(
+            last_hidden, shape=[-1, num_layers, hidden_size], inplace=True)
+        last_hidden = layers.transpose(x=last_hidden, perm=[1, 0, 2])
+
+        last_cell = layers.concat(cell_array, 1)
+        last_cell = layers.reshape(
+            last_cell, shape=[-1, num_layers, hidden_size])
+        last_cell = layers.transpose(x=last_cell, perm=[1, 0, 2])
+
+        real_res = layers.concat(res, 0)
+        real_res = layers.reshape(
+            real_res, shape=[len, -1, hidden_size], inplace=True)
+        real_res = layers.transpose(x=real_res, perm=[1, 0, 2])
+
+        return real_res, last_hidden, last_cell

python/paddle_fl/mobile/model/model_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle.fluid as fluid
+import numpy as np
+
+
+def set_user_param_dict(param_names, param_dict, scope):
+    place = fluid.CPUPlace()
+    for var_name in param_names:
+        param = scope.find_var(var_name)
+        if param is None:
+            print("var name: {} does not exist in memory".format(var_name))
+            continue
+        param.get_tensor().set(param_dict[var_name], place)
+    return
+
+
+def set_global_param_dict(param_names, param_dict, scope):
+    place = fluid.CPUPlace()
+    for var_name in param_names:
+        param = scope.find_var(var_name)
+        if param is None:
+            print("var name: {} does not exist in memory".format(var_name))
+            continue
+        if var_name not in param_dict:
+            print("var name: {} does not exist in global param dict".format(
+                var_name))
+            exit()
+        var_numpy = param_dict[var_name]
+        param.get_tensor().set(var_numpy, place)
+    return
+
+
+class ModelBase(object):
+    def __init__(self):
+        pass
+
+    def init_model(self):
+        pass
+
+    def build_model(self, model_configs):
+        pass
+
+    def get_model_inputs(self):
+        pass
+
+    def get_model_loss(self):
+        pass
+
+    def get_model_metrics(self):
+        pass
+
+    def get_startup_program(self):
+        pass
+
+    def get_main_program(self):
+        pass
+
+    def get_user_param_dict(self):
+        param_dict = {}
+        scope = fluid.global_scope()
+        for var_pair in self.get_user_param_names():
+            param = scope.find_var(var_pair[0])
+            if param is None:
+                print("var name: {} does not exist in memory".format(var_pair[
+                    0]))
+                continue
+            var = param.get_tensor().__array__()
+            param_dict[var_pair[0]] = [var, var_pair[1].shape]
+        return param_dict
+
+    def get_global_param_dict(self):
+        param_dict = {}
+        scope = fluid.global_scope()
+        for var_pair in self.get_global_param_names():
+            param = scope.find_var(var_pair[0])
+            if param is None:
+                print("var name: {} does not exist in memory".format(var_pair[
+                    0]))
+                continue
+            var = param.get_tensor().__array__()
+            param_dict[var_pair[0]] = var
+        return param_dict
+
+    def get_user_param_names(self):
+        user_params = []
+        for var_name, var in self.startup_program_.global_block().vars.items():
+            if var.persistable and "@USER" in var_name and \
+               "learning_rate" not in var_name:
+                user_params.append((var_name, var))
+        return user_params
+
+    def get_global_param_names(self):
+        global_params = []
+        for var_name, var in self.startup_program_.global_block().vars.items():
+            if var.persistable and "@USER" not in var_name and \
+               "learning_rate" not in var_name and "generated_var" not in var_name:
+                global_params.append((var_name, var))
+        return global_params

python/paddle_fl/mobile/optimizer/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .fed_avg_optimizer import FedAvgOptimizer
+from .fed_avg_optimizer import SumOptimizer

python/paddle_fl/mobile/optimizer/fed_avg_optimizer.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .optimizer_base import OptimizerBase
+
+
+class FedAvgOptimizer(OptimizerBase):
+    def __init__(self, learning_rate=1.0):
+        self.learning_rate = learning_rate
+        pass
+
+    def update(self, user_info, new_global_param_by_user, old_global_param,
+               scheduler_client):
+        total_weight = 0.0
+        for uid in user_info:
+            total_weight += float(user_info[uid])
+        update_dict = {}
+        for key in new_global_param_by_user:
+            uid = key
+            uid_global_w = new_global_param_by_user[key]
+            weight = float(user_info[key]) / total_weight
+            for param_name in uid_global_w:
+                if param_name in update_dict:
+                    update_dict[param_name] += \
+                        self.learning_rate * weight * (uid_global_w[param_name] - old_global_param[param_name])
+                else:
+                    update_dict[param_name] = \
+                        self.learning_rate * weight * (uid_global_w[param_name] - old_global_param[param_name])
+
+        scheduler_client.fedavg_update(update_dict)
+
+
+class SumOptimizer(OptimizerBase):
+    def __init__(self, learning_rate=1.0):
+        self.learning_rate = learning_rate
+        pass
+
+    def update(self, user_info, new_global_param_by_user, old_global_param,
+               scheduler_client):
+        update_dict = {}
+        for key in new_global_param_by_user:
+            uid = key
+            uid_global_w = new_global_param_by_user[key]
+            weight = 1.0
+            for param_name in uid_global_w:
+                if param_name in update_dict:
+                    update_dict[param_name] += \
+                        self.learning_rate * weight * (uid_global_w[param_name] - old_global_param[param_name])
+                else:
+                    update_dict[param_name] = \
+                        self.learning_rate * weight * (uid_global_w[param_name] - old_global_param[param_name])
+
+        scheduler_client.fedavg_update(update_dict)

python/paddle_fl/mobile/optimizer/optimizer_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class OptimizerBase(object):
+    def __init__(self, role_maker):
+        pass
+
+    def update(self, new_global_param_by_user, old_global_param,
+               scheduler_client):
+        pass

python/paddle_fl/mobile/protos/data_server.proto

+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+syntax = "proto3";
+package server;
+
+message Data {
+  string date = 1;
+  string uid = 2;
+  string line = 3;
+}
+
+message Res {
+  int32 err_code = 1;
+  int32 user_num = 2;
+}
+
+message UserData {
+  repeated string line_str = 1;
+  int32 err_code = 2;
+}
+
+message Param {
+  string name = 1;
+  repeated float weight = 2;
+  repeated int32 shape = 3;
+}
+
+message UserParams {
+  repeated Param user_params = 1;
+  int32 err_code = 2;
+  string uid = 3;
+}
+
+service DataServer {
+  // (Method definitions not shown)
+  rpc SendData(Data) returns (Res) {}
+  rpc GetUserData(Data) returns (UserData) {}
+  rpc ClearUserData(Data) returns (Res) {}
+  rpc GetUserParams(Data) returns (UserParams) {}
+  rpc UpdateUserParams(UserParams) returns (Res) {}
+}

python/paddle_fl/mobile/protos/run_codegen.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Copyright 2015 gRPC authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Runs protoc with the gRPC plugin to generate messages and gRPC stubs."""
+
+from grpc_tools import protoc
+
+#python -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. data_server.proto
+protoc.main((
+    '',
+    '-I.',
+    '--python_out=.',
+    '--grpc_python_out=.',
+    'data_server.proto', ))
+
+protoc.main((
+    '',
+    '-I.',
+    '--python_out=.',
+    '--grpc_python_out=.',
+    'scheduler_server.proto', ))
+
+import os
+os.system("mv *pb2.py ../servers/")
+os.system("mv *pb2_grpc.py ../servers/")

python/paddle_fl/mobile/protos/scheduler_server.proto

+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+syntax = "proto3";
+package scheduler;
+
+message Request {
+  int32 node_idx = 1;
+  int32 sample_num = 2;
+  int32 shard_num = 3;
+  int32 node_num = 4;
+  int32 min_ins_num = 5;
+  string date = 6;
+}
+
+message UserList {
+  repeated string uids = 1;
+  repeated float weight = 2;
+}
+
+message Param {
+  string name = 1;
+  repeated float weight = 2;
+  repeated int32 shape = 3;
+}
+
+message GlobalParams {
+  repeated Param global_params = 1;
+  int32 err_code = 2;
+}
+
+message UserInstNum {
+  string uid = 1;
+  int32 inst_num = 2;
+}
+
+message UserInstInfo {
+  repeated UserInstNum inst_nums = 1;
+  int32 shard_num = 2;
+  string date = 3;
+}
+
+message Res { int32 err_code = 1; }
+
+message SchedulerServerEmptyInput {}
+
+service SchedulerServer {
+  // (Method definitions not shown)
+  rpc SampleUsersToTrain(Request) returns (UserInstInfo) {}
+  rpc FixedUsersToTrain(Request) returns (UserInstInfo) {}
+  rpc GetGlobalParams(Request) returns (GlobalParams) {}
+  rpc UpdateGlobalParams(GlobalParams) returns (Res) {}
+  rpc FedAvgUpdate(GlobalParams) returns (Res) {}
+  rpc UpdateUserInstNum(UserInstInfo) returns (Res) {}
+  rpc Exit(SchedulerServerEmptyInput) returns (Res) {}
+  rpc SampleUsersToTest(Request) returns (UserInstInfo) {}
+  rpc SampleUsersWithHash(Request) returns (UserInstInfo) {}
+}

python/paddle_fl/mobile/reader/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

python/paddle_fl/mobile/reader/leaf_reddit_reader.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import collections
+import os
+import sys
+import numpy as np
+from utils.logger import logging
+import json
+
+PAD_SYMBOL, UNK_SYMBOL = 0, 1
+DATA_PATH = "lm_data"
+VOCAB_PATH = os.path.join(DATA_PATH, "vocab.json")
+TRAIN_DATA_PATH = os.path.join(DATA_PATH, "20200101", "train_data.json")
+VOCAB = None
+
+
+def build_counter(train_data):
+    train_tokens = []
+    for u in train_data:
+        for c in train_data[u]['x']:
+            train_tokens.extend([s for s in c])
+
+    all_tokens = []
+    for i in train_tokens:
+        all_tokens.extend(i)
+    train_tokens = []
+
+    counter = collections.Counter()
+    counter.update(all_tokens)
+    all_tokens = []
+    return counter
+
+
+def build_vocab(filename, vocab_size=10000):
+    train_data = {}
+    with open(filename) as json_file:
+        data = json.load(json_file)
+        train_data = data['user_data']
+    counter = build_counter(train_data)
+
+    count_pairs = sorted(counter.items(), key=lambda x: (-x[1], x[0]))
+    # -2 to account for the unknown and pad symbols
+    count_pairs = count_pairs[:(vocab_size - 2)]
+    words, _ = list(zip(*count_pairs))
+
+    vocab = {}
+    vocab['<PAD>'] = PAD_SYMBOL
+    vocab['<UNK>'] = UNK_SYMBOL
+
+    for i, w in enumerate(words):
+        if w != '<PAD>':
+            vocab[w] = i + 1
+
+    return {
+        'vocab': vocab,
+        'size': vocab_size,
+        'unk_symbol': vocab['<UNK>'],
+        'pad_symbol': vocab['<PAD>']
+    }
+
+
+def save_vocab(filename, vocab):
+    with open(filename, "w") as f:
+        f.write(json.dumps(vocab))
+
+
+def load_vocab(filename):
+    with open(filename) as f:
+        return json.loads(f.read())
+
+
+if os.path.exists(VOCAB_PATH):
+    logging.info("load vocab form: {}".format(VOCAB_PATH))
+    VOCAB = load_vocab(VOCAB_PATH)
+else:
+    #TODO: singleton
+    logging.info("build vocab form: {}".format(TRAIN_DATA_PATH))
+    VOCAB = build_vocab(TRAIN_DATA_PATH)
+    logging.info("save vocab into: {}".format(VOCAB_PATH))
+    save_vocab(VOCAB_PATH, VOCAB)
+if VOCAB is None:
+    logging.error("load vocab error")
+    raise Exception("load vocab error")
+
+
+def train_reader(lines):
+    def local_iter():
+        seg_id = 0
+        for line in lines:
+            assert (len(line.split("\t")) == 3)
+            uid, _, input_str = line.split("\t")
+            data = json.loads(input_str)
+            data_x = data["x"]
+            data_y = data["y"]
+            data_mask = data["mask"]
+
+            input_data, input_length = process_x(data_x, VOCAB)
+            target_data = process_y(data_y, VOCAB)
+            yield [input_data] + [target_data]
+
+    return local_iter
+
+
+def infer_reader(lines):
+    return train_reader(lines)
+
+
+def load_data_into_patch(filelist, patch_size):
+    data_patch = []
+    idx = 0
+    local_user_dict = {}
+    for fn in filelist:
+        tmp_list = []
+        with open(fn) as fin:
+            raw_data = json.loads(fin.read())["user_data"]
+        local_user_dict = {k: 0 for k in raw_data.keys()}
+        for user, data in raw_data.items():
+            data_x = data["x"]
+            data_y = data["y"]
+            for c, l in zip(data_x, data_y):
+                for inst_i in range(len(c)):
+                    local_user_dict[user] += 1
+                    idx += 1
+                    inst = {
+                        "x": c[inst_i],
+                        "y": l["target_tokens"][inst_i],
+                        "mask": l["count_tokens"][inst_i]
+                    }
+                    line = "{}\t\t{}".format(user, json.dumps(inst))
+                    if idx % patch_size == 0:
+                        data_patch.append(tmp_list)
+                        tmp_list = [line]
+                    else:
+                        tmp_list.append(line)
+        if len(tmp_list) > 0:
+            data_patch.append(tmp_list)
+    return data_patch, local_user_dict
+
+
+def tokens_to_ids(tokens, vocab):
+    to_ret = [vocab.get(word, vocab["<UNK>"]) for word in tokens]
+    return np.array(to_ret, dtype="int64")
+
+
+def process_x(raw_x, vocab):
+    tokens = tokens_to_ids(raw_x, vocab["vocab"])
+    lengths = np.sum(tokens != vocab["pad_symbol"])
+    return tokens, lengths
+
+
+def process_y(raw_y, vocab):
+    tokens = tokens_to_ids(raw_y, vocab["vocab"])
+    return tokens

python/paddle_fl/mobile/sampler/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .uniform_sampler import UniformSampler
+from .fixed_sampler import FixedSampler
+from .sampler_test_user import Test1wSampler
+from .sampler_test_user import Test1percentSampler
+from .hash_sampler import HashSampler

python/paddle_fl/mobile/sampler/fixed_sampler.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .sampler_base import SamplerBase
+
+
+class FixedSampler(SamplerBase):
+    def __init__(self):
+        self.sample_num = 100
+
+    def set_sample_num(self, sample_num):
+        self.sample_num = sample_num
+
+    def sample_user_list(self, scheduler_client, date, sim_idx, shard_num,
+                         sim_num):
+        user_info = scheduler_client.fixed_sample_user_list(
+            date, sim_idx, self.sample_num, shard_num, sim_num)
+        return user_info

python/paddle_fl/mobile/sampler/hash_sampler.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .sampler_base import SamplerBase
+
+
+class HashSampler(SamplerBase):
+    def __init__(self):
+        self.sample_num = 100
+
+    def set_sample_num(self, sample_num):
+        self.sample_num = sample_num
+
+    def sample_user_list(self, scheduler_client, date, sim_idx, shard_num,
+                         sim_num):
+        user_info = scheduler_client.hash_sample_user_list(
+            date, sim_idx, self.sample_num, shard_num, sim_num)
+        return user_info

python/paddle_fl/mobile/sampler/sampler_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class SamplerBase(object):
+    def __init__(self, role_maker):
+        pass
+
+    def sample_user_list(self, scheduler_client, sim_idx, shard_num, sim_num,
+                         min_ins_num):
+        pass

python/paddle_fl/mobile/sampler/sampler_test_user.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .sampler_base import SamplerBase
+
+
+class Test1wSampler(SamplerBase):
+    def __init__(self):
+        pass
+
+    def sample_user_list(self, scheduler_client, date, sim_idx, shard_num,
+                         sim_num):
+        user_info = scheduler_client.sample_test_user_list(date, sim_idx,
+                                                           shard_num, sim_num)
+        return user_info
+
+
+class Test1percentSampler(SamplerBase):
+    def __init__(self):
+        pass
+
+    def sample_user_list(self, scheduler_client, date, sim_idx, shard_num,
+                         sim_num):
+        user_info = scheduler_client.sample_test_user_list(date, sim_idx,
+                                                           shard_num, sim_num)
+        return user_info

python/paddle_fl/mobile/sampler/uniform_sampler.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .sampler_base import SamplerBase
+
+
+class UniformSampler(SamplerBase):
+    def __init__(self):
+        self.sample_num = 100
+        self.min_ins_num = 1
+
+    def set_sample_num(self, sample_num):
+        self.sample_num = sample_num
+
+    def set_min_ins_num(self, min_ins_num):
+        self.min_ins_num = min_ins_num
+
+    def sample_user_list(self, scheduler_client, date, sim_idx, shard_num,
+                         sim_num):
+        user_info = scheduler_client.uniform_sample_user_list(
+            date, sim_idx, self.sample_num, shard_num, sim_num,
+            self.min_ins_num)
+        return user_info

python/paddle_fl/mobile/servers/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .data_server_impl import DataServerServicer
+from .data_server_impl import DataServer
+from .scheduler_server_impl import SchedulerServerServicer
+from .scheduler_server_impl import SchedulerServer

python/paddle_fl/mobile/servers/data_server_impl.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+from concurrent import futures
+import data_server_pb2
+import data_server_pb2_grpc
+import grpc
+import sys
+from utils.logger import logging
+
+
+class DataServerServicer(object):
+    def __init__(self):
+        self.data_dict = {}
+        self.param_dict = {}
+        self.request_num = 0
+
+    def GetUserParams(self, request, context):
+        uid = unicode(request.uid)
+        if uid in self.param_dict:
+            return self.param_dict[uid]
+        else:
+            user_param = data_server_pb2.UserParams()
+            user_param.err_code = -1
+            return user_param
+
+    def UpdateUserParams(self, request, context):
+        self.param_dict[request.uid] = request
+        res = data_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+    def ClearUserData(self, request, context):
+        date = request.date
+        self.data_dict[date].clear()
+        res = data_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+    def GetUserData(self, request, context):
+        uid = unicode(request.uid)
+        date = unicode(request.date)
+        if date not in self.data_dict:
+            user_data = data_server_pb2.UserData()
+            user_data.err_code = -1
+            return user_data
+        if uid in self.data_dict[date]:
+            self.data_dict[date][uid].err_code = 0
+            return self.data_dict[date][uid]
+        else:
+            user_data = data_server_pb2.UserData()
+            user_data.err_code = -1
+            self.data_dict[date][uid] = user_data
+            return self.data_dict[date][uid]
+        user_data = data_server_pb2.UserData()
+        user_data.err_code = -1
+        return user_data
+
+    def SendData(self, request, context):
+        date = unicode(request.date)
+        uid = unicode(request.uid)
+        if date in self.data_dict:
+            if uid in self.data_dict[request.date]:
+                self.data_dict[date][uid].line_str.extend([request.line])
+            else:
+                user_data = data_server_pb2.UserData()
+                self.data_dict[date][uid] = user_data
+                self.data_dict[date][uid].line_str.extend([request.line])
+        else:
+            user_data = data_server_pb2.UserData()
+            self.data_dict[date] = {}
+            self.data_dict[date][uid] = user_data
+            self.data_dict[date][uid].line_str.extend([request.line])
+
+        res = data_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+
+class DataServer(object):
+    def __init__(self):
+        pass
+
+    def start(self, max_workers=1000, concurrency=100, endpoint=""):
+        if endpoint == "":
+            logging.error("You should specify endpoint in start function")
+            return
+        server = grpc.server(
+            futures.ThreadPoolExecutor(max_workers=max_workers),
+            options=[('grpc.max_send_message_length', 1024 * 1024 * 1024),
+                     ('grpc.max_receive_message_length', 1024 * 1024 * 1024)],
+            maximum_concurrent_rpcs=concurrency)
+        data_server_pb2_grpc.add_DataServerServicer_to_server(
+            DataServerServicer(), server)
+        server.add_insecure_port('[::]:{}'.format(endpoint))
+        server.start()
+        server.wait_for_termination()
+
+
+if __name__ == "__main__":
+    data_server = DataServer()
+    data_server.start(endpoint=sys.argv[1])

python/paddle_fl/mobile/servers/scheduler_server_impl.py

+# -*- coding: utf-8 -*-
+from __future__ import print_function
+from concurrent import futures
+import scheduler_server_pb2
+import scheduler_server_pb2_grpc
+import grpc
+import numpy as np
+import random
+import sys, os
+import time
+import xxhash
+from utils.logger import logging
+
+
+class SchedulerServerServicer(object):
+    def __init__(self):
+        self.global_param_dict = {}
+        self.uid_inst_num_dict = {}
+        self.shard_id_dict = {}
+
+    def uid_shard(self, uid, shard_num):
+        try:
+            # print("uid_shard uid: %s" % uid)
+            uid_hash = xxhash.xxh32(str(uid), seed=101).intdigest()
+        except:
+            return -1
+        shard_idx = uid_hash % shard_num
+        return shard_idx
+
+    def is_test_uid(self, uid):
+        return xxhash.xxh32(str(uid), seed=222).intdigest() % 100 == 3
+
+    # we suppose shard num will not be changed during one training job
+    # but can be changed with another job
+    # so we send shard num every time we update user inst num
+    def UpdateUserInstNum(self, request, context):
+        shard_num = request.shard_num
+        date = request.date
+        if date not in self.uid_inst_num_dict:
+            self.uid_inst_num_dict[date] = {}
+        if date not in self.shard_id_dict:
+            self.shard_id_dict[date] = {}
+        for user in request.inst_nums:
+            shard_id = self.uid_shard(user.uid, shard_num)
+            if shard_id == -1:
+                logging.info("UpdateUserInstNum continue")
+                continue
+            if user.uid in self.uid_inst_num_dict[date]:
+                self.uid_inst_num_dict[date][user.uid] += user.inst_num
+            else:
+                self.uid_inst_num_dict[date][user.uid] = user.inst_num
+            if shard_id not in self.shard_id_dict[date]:
+                self.shard_id_dict[date][shard_id] = [user.uid]
+            else:
+                self.shard_id_dict[date][shard_id].append(user.uid)
+        res = scheduler_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+    '''
+    SampleUsersToTrain: 
+        request.node_idx: from which worker node the request is from
+        request.sample_num: how many users do we need to sample
+        request.shard_num: total shard number of this task
+        request.node_num: total number of training node
+    '''
+
+    def SampleUsersToTrain(self, request, context):
+        node_idx = request.node_idx
+        sample_num = request.sample_num
+        shard_num = request.shard_num
+        node_num = request.node_num
+        date = request.date
+        shard_per_node = shard_num / node_num
+        begin_idx = node_idx * shard_per_node
+        min_ins_num = request.min_ins_num
+
+        uid_list = []
+        i = 0
+        while i < sample_num:
+            shard_idx = begin_idx + random.randint(0, shard_per_node)
+            if shard_idx not in self.shard_id_dict[date]:
+                continue
+            sample_idx = random.randint(
+                0, len(self.shard_id_dict[date][shard_idx]) - 1)
+            uid = self.shard_id_dict[date][shard_idx][sample_idx]
+            if self.uid_inst_num_dict[date][uid] < min_ins_num:
+                continue
+            uid_list.append(uid)
+            i += 1
+
+        info = scheduler_server_pb2.UserInstInfo()
+        for uid in uid_list:
+            inst_num = scheduler_server_pb2.UserInstNum()
+            inst_num.uid = uid
+            inst_num.inst_num = self.uid_inst_num_dict[date][uid]
+            info.inst_nums.extend([inst_num])
+        return info
+
+    def SampleUsersWithHash(self, request, context):
+        node_idx = request.node_idx
+        sample_num = request.sample_num
+        shard_num = request.shard_num
+        node_num = request.node_num
+        date = request.date
+        shard_per_node = shard_num / node_num
+        begin_idx = node_idx * shard_per_node
+
+        uid_list = []
+        i = 0
+        while i < sample_num:
+            shard_idx = begin_idx + random.randint(0, shard_per_node)
+            if shard_idx not in self.shard_id_dict[date]:
+                continue
+            sample_idx = random.randint(
+                0, len(self.shard_id_dict[date][shard_idx]) - 1)
+            uid = self.shard_id_dict[date][shard_idx][sample_idx]
+            if not self.is_test_uid(uid):
+                continue
+            uid_list.append(uid)
+            i += 1
+
+        info = scheduler_server_pb2.UserInstInfo()
+        for uid in uid_list:
+            inst_num = scheduler_server_pb2.UserInstNum()
+            inst_num.uid = uid
+            inst_num.inst_num = self.uid_inst_num_dict[date][uid]
+            info.inst_nums.extend([inst_num])
+        return info
+
+    def SampleUsersToTest(self, request, context):
+        node_idx = request.node_idx
+        shard_num = request.shard_num
+        node_num = request.node_num
+        date = request.date
+        shard_per_node = shard_num / node_num
+        shard_begin_idx = node_idx * shard_per_node
+        shard_end_idx = (node_idx + 1) * shard_per_node
+        uid_list = []
+
+        for shard_idx in range(shard_begin_idx, shard_end_idx):
+            for uid in self.shard_id_dict[date][shard_idx]:
+                if self.is_test_uid(uid):
+                    uid_list.append(uid)
+        info = scheduler_server_pb2.UserInstInfo()
+        for uid in uid_list:
+            inst_num = scheduler_server_pb2.UserInstNum()
+            inst_num.uid = uid
+            inst_num.inst_num = self.uid_inst_num_dict[date][uid]
+            info.inst_nums.extend([inst_num])
+        return info
+
+    def FixedUsersToTrain(self, request, context):
+        node_idx = request.node_idx
+        sample_num = request.sample_num
+        shard_num = request.shard_num
+        node_num = request.node_num
+        date = request.date
+        begin_idx = node_idx * shard_num
+
+        shard_per_node = shard_num / node_num
+
+        uid_list = []
+        i = 0
+        assert (sample_num <= 100)
+        with open("data/test_user_100.txt") as f:
+            for line in f.readlines():
+                uid_list.append(line.strip())
+        # uid_list.extend(["somebody", "nobody"])
+        info = scheduler_server_pb2.UserInstInfo()
+        for uid in uid_list[:sample_num]:
+            inst_num = scheduler_server_pb2.UserInstNum()
+            inst_num.uid = uid
+            inst_num.inst_num = self.uid_inst_num_dict[date][uid]
+            info.inst_nums.extend([inst_num])
+        return info
+
+    def GetGlobalParams(self, request, context):
+        if self.global_param_dict == {}:
+            logging.debug("global param has not been initialized")
+            return
+        # logging.info("node {} is asking for global params".format(request.node_idx))
+        global_param_pb = scheduler_server_pb2.GlobalParams()
+        for key in self.global_param_dict:
+            param = scheduler_server_pb2.Param()
+            param.name = key
+            var, shape = self.global_param_dict[key]
+            param.weight.extend(var)
+            # print("GetGlobalParams before param.shape")
+            param.shape.extend(shape)
+            # print("GetGlobalParams after param.shape")
+            global_param_pb.global_params.extend([param])
+        # print("finish GetGlobalParams")
+        return global_param_pb
+
+    def UpdateGlobalParams(self, request, context):
+        for param in request.global_params:
+            self.global_param_dict[param.name] = [param.weight, param.shape]
+        res = scheduler_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+    def FedAvgUpdate(self, request, context):
+        for param in request.global_params:
+            old_param, shape = self.global_param_dict[param.name]
+            for idx, item in enumerate(old_param):
+                old_param[idx] += param.weight[idx]
+        res = scheduler_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+    def Exit(self, request, context):
+        with open("_shutdown_scheduler", "w") as f:
+            f.write("_shutdown_scheduler\n")
+        res = scheduler_server_pb2.Res()
+        res.err_code = 0
+        return res
+
+
+class SchedulerServer(object):
+    def __init__(self):
+        pass
+
+    def start(self, max_workers=1000, concurrency=100, endpoint=""):
+        if endpoint == "":
+            logging.info("You should specify endpoint in start function")
+            return
+        server = grpc.server(
+            futures.ThreadPoolExecutor(max_workers=max_workers),
+            options=[('grpc.max_send_message_length', 1024 * 1024 * 1024),
+                     ('grpc.max_receive_message_length', 1024 * 1024 * 1024)],
+            maximum_concurrent_rpcs=concurrency)
+        scheduler_server_pb2_grpc.add_SchedulerServerServicer_to_server(
+            SchedulerServerServicer(), server)
+        # print("SchedulerServer add endpoint: ", '[::]:{}'.format(endpoint))
+        server.add_insecure_port('[::]:{}'.format(endpoint))
+        server.start()
+        logging.info("server started")
+        os.system("rm _shutdown_scheduler")
+        while (not os.path.isfile("_shutdown_scheduler")):
+            time.sleep(10)
+
+
+if __name__ == "__main__":
+    scheduler_server = SchedulerServer()
+    scheduler_server.start(endpoint=60001)

python/paddle_fl/mobile/trainer/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .language_model_trainer import LanguageModelTrainer

python/paddle_fl/mobile/trainer/language_model_trainer.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .trainer_base import TrainerBase
+from model import LanguageModel
+from clients import DataClient
+import paddle.fluid as fluid
+from utils.hdfs_utils import multi_upload, HDFSClient
+import reader.leaf_reddit_reader as reader
+from utils.logger import logging
+from itertools import groupby
+import numpy as np
+import random
+import paddle
+import pickle
+import os
+from model.model_base import set_user_param_dict
+from model.model_base import set_global_param_dict
+
+
+def train_one_user(arg_dict, trainer_config):
+    show_metric = trainer_config["show_metric"]
+    shuffle = trainer_config["shuffle"]
+    max_training_steps = trainer_config["max_training_steps"]
+    batch_size = trainer_config["batch_size"]
+    # logging.info("training one user...")
+    main_program = fluid.Program.parse_from_string(trainer_config[
+        "main_program_desc"])
+    startup_program = fluid.Program.parse_from_string(trainer_config[
+        "startup_program_desc"])
+    place = fluid.CPUPlace()
+    exe = fluid.Executor(place)
+    scope = fluid.global_scope()
+    if (startup_program is None):
+        logging.error("startup_program is None")
+        exit()
+    exe.run(startup_program)
+
+    feeder = fluid.DataFeeder(
+        feed_list=trainer_config["input_names"],
+        place=place,
+        program=main_program)
+    data_server_endpoints = arg_dict["data_endpoints"]
+    # create data clients
+    data_client = DataClient()
+    data_client.set_data_server_endpoints(data_server_endpoints)
+    uid = arg_dict["uid"]
+    date = arg_dict["date"]
+    global_param_dict = arg_dict["global_params"]
+    user_data = data_client.get_data_by_uid(uid, date)
+    train_reader = reader.train_reader(user_data)
+    if shuffle == True:
+        train_reader = paddle.reader.shuffle(train_reader, buf_size=10000)
+    train_reader = paddle.batch(train_reader, batch_size=batch_size)
+
+    # get user param
+    # logging.debug("do not need to get user params")
+
+    set_global_param_dict(arg_dict["global_param_names"],
+                          arg_dict["global_params"], scope)
+
+    if (main_program is None):
+        logging.error("main_program is None")
+        exit()
+
+    epoch = trainer_config["epoch"]
+    max_steps_in_epoch = trainer_config.get("max_steps_in_epoch", -1)
+    metrics = trainer_config["metrics"]
+    metric_keys = metrics.keys()
+    fetch_list = [main_program.global_block().var(trainer_config["loss_name"])]
+    for key in metric_keys:
+        fetch_list.append(main_program.global_block().var(metrics[key]))
+
+    seq_len = 10
+    for ei in range(epoch):
+        trained_sample_num = 0
+        step = 0
+        fetch_res_list = []
+        total_loss = 0.0
+        total_correct = 0
+        for data in train_reader():
+            fetch_res = exe.run(main_program,
+                                feed=feeder.feed(data),
+                                fetch_list=fetch_list)
+            step += 1
+            trained_sample_num += len(data)
+            fetch_res_list.append([x[0] for x in fetch_res])
+            if max_steps_in_epoch != -1 and step >= max_steps_in_epoch:
+                break
+
+        if show_metric and trained_sample_num > 0:
+            loss = sum([x[0] for x in fetch_res_list]) / trained_sample_num
+            print("loss: {}, ppl: {}".format(loss, np.exp(loss)))
+            for i, key in enumerate(metric_keys):
+                if key == "correct":
+                    value = float(sum([x[i + 1] for x in fetch_res_list
+                                       ])) / trained_sample_num
+                    print("correct: {}".format(value / seq_len))
+
+    local_updated_param_dict = {}
+    # update user param
+    # logging.debug("do not need to update user params")
+
+    data_client.set_param_by_uid(uid, local_updated_param_dict)
+    # global_updated_param_dict = {}
+    write_global_param_file = arg_dict["write_global_param_file"]
+    #os.makedirs("%s/params" % write_global_param_file)
+    for var_name in arg_dict["global_param_names"]:
+        var = scope.var(var_name).get_tensor().__array__().astype(np.float32)
+        filename = os.path.join(write_global_param_file, "params", var_name)
+        #logging.info("filename: {}".format(filename))
+        dirname = os.path.dirname(filename)
+        if not os.path.exists(dirname):
+            os.makedirs(dirname)
+        with open(filename, "w") as f:
+            np.save(f, var)
+    with open("%s/_info" % write_global_param_file, "w") as f:
+        pickle.dump([uid, trained_sample_num], file=f)
+
+
+def infer_one_user(arg_dict, trainer_config):
+    """
+    infer a model with global_param and user params
+    input:
+        global_param
+        user_params
+        infer_program
+        user_data
+    output:
+        [sample_cout, top1] 
+    """
+    # run startup program, set params
+    uid = arg_dict["uid"]
+    batch_size = trainer_config["batch_size"]
+    startup_program = fluid.Program.parse_from_string(trainer_config[
+        "startup_program_desc"])
+    infer_program = fluid.Program.parse_from_string(trainer_config[
+        "infer_program_desc"])
+    place = fluid.CPUPlace()
+    exe = fluid.Executor(place)
+    scope = fluid.global_scope()
+
+    if (startup_program is None):
+        logging.error("startup_program is None")
+        exit()
+    if (infer_program is None):
+        logging.error("infer_program is None")
+        exit()
+
+    exe.run(startup_program)
+
+    data_client = DataClient()
+    data_client.set_data_server_endpoints(arg_dict["data_endpoints"])
+
+    # get user param
+    # logging.debug("do not need to get user params")
+
+    set_global_param_dict(arg_dict["global_param_names"],
+                          arg_dict["global_params"], scope)
+
+    # reader
+
+    date = arg_dict["date"]
+    global_param_dict = arg_dict["global_params"]
+    user_data = data_client.get_data_by_uid(uid, date)
+    infer_reader = reader.infer_reader(user_data)
+    infer_reader = paddle.batch(infer_reader, batch_size=batch_size)
+
+    # run infer program
+    os.mkdir(arg_dict["infer_result_dir"])
+    #pred_file = open(arg_dict["infer_result_dir"] + '/' + "pred_file", "w")
+    feeder = fluid.DataFeeder(
+        feed_list=trainer_config["input_names"],
+        place=place,
+        program=infer_program)
+
+    fetch_list = trainer_config["target_names"]
+    #logging.info("fetch_list: {}".format(fetch_list))
+    fetch_res = []
+    sample_count = 0
+
+    total_loss = 0.0
+    total_correct = 0
+    iters = 0
+    steps = 0
+    seq_len = 10
+    for data in infer_reader():
+        # feed_data = [x["features"] + [x["label"]] for x in data]
+        # prediction, acc_val= exe.run(infer_program,
+        pred, correct_count, loss = exe.run(infer_program,
+                                            feed=feeder.feed(data),
+                                            fetch_list=fetch_list)
+        total_loss += loss
+        total_correct += correct_count
+        steps += 1
+        sample_count += len(data)
+
+    correct = float(total_correct) / (seq_len * sample_count)
+    # logging.info("correct: {}".format(correct))
+    with open(arg_dict["infer_result_dir"] + "/res", "w") as f:
+        f.write("%d\t%f\n" % (1, correct))
+
+
+def save_and_upload(arg_dict, trainer_config, dfs_upload_path):
+    logging.info("do not save and upload...")
+    return
+
+
+def evaluate_a_group(group):
+    group_list = []
+    for label, pred, _ in group:
+        # print("%s\t%s\n" % (label, pred))
+        group_list.append((int(label), float(pred)))
+    random.shuffle(group_list)
+    labels = [x[0] for x in group_list]
+    preds = [x[1] for x in group_list]
+    true_res = labels.index(1) if 1 in labels else -1
+    pred_res = preds.index(max(preds))
+    if pred_res == true_res:
+        return 1
+    else:
+        return 0
+
+
+class LanguageModelTrainer(TrainerBase):
+    """
+    LanguageModelTrainer only support training with PaddlePaddle
+    """
+
+    def __init__(self):
+        super(LanguageModelTrainer, self).__init__()
+        self.main_program_ = fluid.Program()
+        self.startup_program_ = fluid.Program()
+        self.infer_program_ = fluid.Program()
+        self.main_program_desc_ = ""
+        self.startup_program_desc_ = ""
+        self.infer_program_desc_ = ""
+        self.train_one_user_func = train_one_user
+        self.infer_one_user_func = infer_one_user
+        self.save_and_upload_func = save_and_upload
+        self.input_model_ = None
+
+    def get_load_data_into_patch_func(self):
+        return reader.load_data_into_patch
+
+    def prepare(self, do_test=False):
+        self.generate_program_desc(do_test)
+        pass
+
+    def get_user_param_names(self):
+        # return [x[0] for x in self.input_model_.get_user_param_names()]
+        pass
+
+    def get_global_param_names(self):
+        return [x[0] for x in self.input_model_.get_global_param_names()]
+
+    def generate_program_desc(self, do_test=False):
+        """
+        generate the paddle program desc
+        """
+        with fluid.program_guard(self.main_program_, self.startup_program_):
+            self.input_model_ = LanguageModel()
+            model_configs = {}
+            self.input_model_.build_model(model_configs)
+            optimizer = fluid.optimizer.SGD(
+                learning_rate=self.trainer_config["lr"])
+            optimizer.minimize(self.input_model_.get_model_loss())
+
+        self.main_program_desc_ = self.main_program_.desc.serialize_to_string()
+        self.startup_program_desc_ = self.startup_program_.desc.serialize_to_string(
+        )
+        self.update_trainer_configs("loss_name",
+                                    self.input_model_.get_model_loss_name())
+        self.update_trainer_configs(
+            "input_names",
+            self.input_model_.get_model_input_names(), )
+        self.update_trainer_configs(
+            "target_names",
+            self.input_model_.get_target_names(), )
+        self.update_trainer_configs(
+            "metrics",
+            self.input_model_.get_model_metrics(), )
+        self.update_trainer_configs("show_metric", True)
+        self.update_trainer_configs("max_training_steps", "inf")
+        self.update_trainer_configs("shuffle", False)
+        self.update_trainer_configs("main_program_desc",
+                                    self.main_program_desc_)
+        self.update_trainer_configs("startup_program_desc",
+                                    self.startup_program_desc_)
+
+        if do_test:
+            input_names = self.input_model_.get_model_input_names()
+            target_var_names = self.input_model_.get_target_names()
+            self.infer_program_ = self.main_program_._prune_with_input(
+                feeded_var_names=input_names, targets=target_var_names)
+            self.infer_program_ = self.infer_program_._inference_optimize(
+                prune_read_op=True)
+            fluid.io.prepend_feed_ops(self.infer_program_, input_names)
+            fluid.io.append_fetch_ops(self.infer_program_, target_var_names)
+            self.infer_program_.desc._set_version()
+            fluid.core.save_op_compatible_info(self.infer_program_.desc)
+            self.infer_program_desc_ = self.infer_program_.desc.serialize_to_string(
+            )
+            self.update_trainer_configs("infer_program_desc",
+                                        self.infer_program_desc_)
+
+    def init_global_model(self, scheduler_client):
+        logging.info("initializing global model")
+        place = fluid.CPUPlace()
+        exe = fluid.Executor(place)
+        exe.run(self.startup_program_)
+        logging.info("finish initializing global model")
+
+        global_param_dict = self.input_model_.get_global_param_dict()
+        scheduler_client.update_global_params(global_param_dict)

python/paddle_fl/mobile/trainer/trainer_base.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class TrainerBase(object):
+    def __init__(self):
+        self.trainer_config = None
+        self.train_one_user_func = None
+        self.infer_one_user_func = None
+        self.save_and_upload_func = None
+
+    def get_user_param_names(self):
+        pass
+
+    def get_global_param_names(self):
+        pass
+
+    def set_trainer_configs(self, trainer_configs):
+        """
+        config training parameter, only support the basic types of python
+        """
+        self.trainer_config = trainer_configs
+
+    def update_trainer_configs(self, key, val):
+        self.trainer_config[key] = val
+
+    def prepare(self):
+        """
+        generate network description string;
+        """
+        pass
+
+    def init_global_model(self, scheduler_client):
+        """
+        initialize the network parameters, which will be broadcasted to all simulator
+        """
+        pass

python/paddle_fl/mobile/utils/__init__.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .role_maker import FLSimRoleMaker

python/paddle_fl/mobile/utils/logger.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import logging
+
+logging.basicConfig(
+    stream=sys.stdout,
+    format='%(asctime)s  %(filename)s : %(levelname)s  %(message)s',
+    level=logging.DEBUG)
+logger = logging.getLogger(__name__)

python/paddle_fl/mobile/utils/role_maker.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from utils.logger import logging
+
+
+class FLSimRoleMaker(object):
+    def __init__(self):
+        from mpi4py import MPI
+        self.MPI = MPI
+        self._comm = MPI.COMM_WORLD
+
+    def init_env(self, local_shard_num=1):
+        if (local_shard_num > 250):
+            logging.error("shard num must be less than or equal to 250")
+            exit()
+        self.free_ports = self._get_free_endpoints(local_shard_num, 40000)
+        import socket
+        ip = socket.gethostbyname(socket.gethostname())
+        self.free_endpoints = ["{}:{}".format(ip, x) for x in self.free_ports]
+        self._comm.barrier()
+        self.node_type = 1
+        if self._comm.Get_rank() == 0:
+            self.node_type = 0
+        self.group_comm = self._comm.Split(self.node_type)
+        self.all_endpoints = self._comm.allgather(self.free_endpoints)
+
+    def simulator_num(self):
+        return self._comm.Get_size() - 1
+
+    def simulator_idx(self):
+        return self._comm.Get_rank() - 1
+
+    def get_global_scheduler_endpoint(self):
+        return self.all_endpoints[0][0]
+
+    def get_data_server_endpoints(self):
+        #return self.all_endpoints[2:][::2]
+        all_endpoints = []
+        for eps in self.all_endpoints[1:]:
+            all_endpoints.extend(eps[:len(eps) / 2])
+        return all_endpoints
+
+    def get_local_data_server_endpoint(self):
+        if self._comm.Get_rank() < 1:
+            return None
+        local_endpoints = self.all_endpoints[self._comm.Get_rank()]
+        return local_endpoints[:len(local_endpoints) / 2]
+
+    def get_local_param_server_endpoint(self):
+        if self._comm.Get_rank() < 1:
+            return None
+        local_endpoints = self.all_endpoints[self._comm.Get_rank()]
+        return local_endpoints[len(local_endpoints) / 2:]
+
+    def is_global_scheduler(self):
+        rank = self._comm.Get_rank()
+        return rank == 0
+
+    def is_simulator(self):
+        return self._comm.Get_rank() > 0
+
+    def barrier_simulator(self):
+        if self._comm.Get_rank() > 0:
+            self.group_comm.barrier()
+
+    def barrier(self):
+        self.group_comm.barrier()
+
+    def _get_free_endpoints(self, local_shard_num, start_port):
+        import psutil
+        conns = psutil.net_connections()
+        x = [conn.laddr.port for conn in conns]
+        free_endpoints = []
+        step = 500
+        start_range = start_port + self._comm.Get_rank() * step
+        for i in range(start_range, start_range + step, 1):
+            if i in x:
+                continue
+            if i > 65535:
+                continue
+            else:
+                free_endpoints.append(i)
+            if len(free_endpoints) == local_shard_num * 2:
+                break
+        return free_endpoints

python/paddle_fl/mobile/utils/run_servers.sh

+endpoint=("50050" "50051" "50052" "50053" "50054" "50055" "50056" "50057" "50058" "50059")
+
+for i in {0..9}
+do
+    python data_server_impl.py ${endpoint[$i]} &
+done

python/paddle_fl/mobile/utils/test_role_maker.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from role_maker import FLSimRoleMaker
+
+role_maker = FLSimRoleMaker()
+role_maker.init_env(local_shard_num=30)
+print("simulator num: {}".format(role_maker.simulator_num()))
+print("simulator idx: {}".format(role_maker.simulator_idx()))
+print("global scheduler endpoint: {}".format(
+    role_maker.get_global_scheduler_endpoint()))
+print("data server endpoints")
+print(role_maker.get_data_server_endpoints())
+print("local data server")
+print(role_maker.get_local_data_server_endpoint())
+print("local param server")
+print(role_maker.get_local_param_server_endpoint())
+role_maker.barrier_simulator()