Merge remote-tracking branch 'chengmo/fix_setup_print' into fix_setup_print

core/trainers/framework/dataset.py

@@ -15,13 +15,13 @@
 from __future__ import print_function
 
 import os
-import warnings
 
 import paddle.fluid as fluid
 from paddlerec.core.utils import envs
 from paddlerec.core.utils import dataloader_instance
 from paddlerec.core.reader import SlotReader
 from paddlerec.core.trainer import EngineMode
+from paddlerec.core.utils.util import split_files
 
 __all__ = ["DatasetBase", "DataLoader", "QueueDataset"]
 
@@ -123,7 +123,8 @@ class QueueDataset(DatasetBase):
             for x in os.listdir(train_data_path)
         ]
         if context["engine"] == EngineMode.LOCAL_CLUSTER:
-            file_list = context["fleet"].split_files(file_list)
+            file_list = split_files(file_list, context["fleet"].worker_index(),
+                                    context["fleet"].worker_num())
 
         dataset.set_filelist(file_list)
         for model_dict in context["phases"]:

core/utils/dataloader_instance.py

@@ -19,6 +19,7 @@ from paddlerec.core.utils.envs import get_global_env
 from paddlerec.core.utils.envs import get_runtime_environ
 from paddlerec.core.reader import SlotReader
 from paddlerec.core.trainer import EngineMode
+from paddlerec.core.utils.util import split_files
 
 
 def dataloader_by_name(readerclass,
@@ -39,7 +40,8 @@ def dataloader_by_name(readerclass,
 
     files = [str(data_path) + "/%s" % x for x in os.listdir(data_path)]
     if context["engine"] == EngineMode.LOCAL_CLUSTER:
-        files = context["fleet"].split_files(files)
+        files = split_files(files, context["fleet"].worker_index(),
+                            context["fleet"].worker_num())
         print("file_list : {}".format(files))
 
     reader = reader_class(yaml_file)
@@ -80,7 +82,8 @@ def slotdataloader_by_name(readerclass, dataset_name, yaml_file, context):
 
     files = [str(data_path) + "/%s" % x for x in os.listdir(data_path)]
     if context["engine"] == EngineMode.LOCAL_CLUSTER:
-        files = context["fleet"].split_files(files)
+        files = split_files(files, context["fleet"].worker_index(),
+                            context["fleet"].worker_num())
         print("file_list: {}".format(files))
 
     sparse = get_global_env(name + "sparse_slots", "#")
@@ -133,7 +136,8 @@ def slotdataloader(readerclass, train, yaml_file, context):
 
     files = [str(data_path) + "/%s" % x for x in os.listdir(data_path)]
     if context["engine"] == EngineMode.LOCAL_CLUSTER:
-        files = context["fleet"].split_files(files)
+        files = split_files(files, context["fleet"].worker_index(),
+                            context["fleet"].worker_num())
         print("file_list: {}".format(files))
 
     sparse = get_global_env("sparse_slots", "#", namespace)

core/utils/envs.py

@@ -18,7 +18,9 @@ import copy
 import os
 import socket
 import sys
+import six
 import traceback
+import six
 
 global_envs = {}
 global_envs_flatten = {}
@@ -101,6 +103,12 @@ def set_global_envs(envs):
             name = ".".join(["dataset", dataset["name"], "type"])
             global_envs[name] = "DataLoader"
 
+    if get_platform() == "LINUX" and six.PY3:
+        print("QueueDataset can not support PY3, change to DataLoader")
+        for dataset in envs["dataset"]:
+            name = ".".join(["dataset", dataset["name"], "type"])
+            global_envs[name] = "DataLoader"
+
 
 def get_global_env(env_name, default_value=None, namespace=None):
     """
@@ -253,11 +261,19 @@ def load_yaml(config):
             use_full_loader = False
 
     if os.path.isfile(config):
-        with open(config, 'r') as rb:
-            if use_full_loader:
-                _config = yaml.load(rb.read(), Loader=yaml.FullLoader)
-            else:
-                _config = yaml.load(rb.read())
-            return _config
+        if six.PY2:
+            with open(config, 'r') as rb:
+                if use_full_loader:
+                    _config = yaml.load(rb.read(), Loader=yaml.FullLoader)
+                else:
+                    _config = yaml.load(rb.read())
+                return _config
+        else:
+            with open(config, 'r', encoding="utf-8") as rb:
+                if use_full_loader:
+                    _config = yaml.load(rb.read(), Loader=yaml.FullLoader)
+                else:
+                    _config = yaml.load(rb.read())
+                return _config
     else:
         raise ValueError("config {} can not be supported".format(config))

core/utils/util.py

@@ -19,11 +19,8 @@ import time
 import numpy as np
 from paddle import fluid
 
-from paddlerec.core.utils import fs as fs
-
 
 def save_program_proto(path, program=None):
-
     if program is None:
         _program = fluid.default_main_program()
     else:
@@ -171,6 +168,39 @@ def print_cost(cost, params):
     return log_str
 
 
+def split_files(files, trainer_id, trainers):
+    """
+    split files before distributed training,
+    example 1: files is [a, b, c ,d, e]  and trainer_num = 2, then trainer
+               0 gets [a, b, c] and trainer 1 gets [d, e].
+    example 2: files is [a, b], and trainer_num = 3, then trainer 0 gets
+               [a], trainer 1 gets [b],  trainer 2 gets []
+
+    Args:
+        files(list): file list need to be read.
+
+    Returns:
+        list: files belongs to this worker.
+    """
+    if not isinstance(files, list):
+        raise TypeError("files should be a list of file need to be read.")
+
+    remainder = len(files) % trainers
+    blocksize = int(len(files) / trainers)
+
+    blocks = [blocksize] * trainers
+    for i in range(remainder):
+        blocks[i] += 1
+
+    trainer_files = [[]] * trainers
+    begin = 0
+    for i in range(trainers):
+        trainer_files[i] = files[begin:begin + blocks[i]]
+        begin += blocks[i]
+
+    return trainer_files[trainer_id]
+
+
 class CostPrinter(object):
     """
     For count cost time && print cost log

models/contentunderstanding/readme.md

@@ -86,7 +86,7 @@ python -m paddlerec.run -m ./config.yaml # 自定义修改超参后，指定配
 # 修改对应模型的config.yaml, workspace配置为当前目录的绝对路径
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: train_runner -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 修改phase阶段为infer的配置，参照config注释
 
 # 修改完config.yaml后 执行:
@@ -106,7 +106,7 @@ python -m paddlerec.run -m ./config.yaml # 自定义修改超参后，指定配
 # 修改对应模型的config.yaml, workspace配置为当前目录的绝对路径
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: train_runner -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 修改phase阶段为infer的配置，参照config注释
 
 # 修改完config.yaml后 执行:

models/demo/movie_recommand/rank/config.yaml

@@ -64,8 +64,7 @@ runner:
   device: cpu
 
 - name: runner_infer
-  epochs: 1
-  class: single_infer
+  class: infer
   print_interval: 10000
   init_model_path: "increment/9" # load model path
 

models/demo/movie_recommand/recall/config.yaml

@@ -64,8 +64,7 @@ runner:
   device: cpu
 
 - name: runner_infer
-  epochs: 1
-  class: single_infer
+  class: infer
   print_interval: 10000
   init_model_path: "increment/9" # load model path
 

models/match/dssm/config.yaml

@@ -56,9 +56,7 @@ runner:
   init_model_path: "" # load model path
   print_interval: 2
 - name: infer_runner
-  class: single_infer
-  # num of epochs
-  epochs: 1
+  class: infer
   # device to run training or infer
   device: cpu
   print_interval: 1

models/match/multiview-simnet/config.yaml

@@ -63,9 +63,7 @@ runner:
   init_model_path: "" # load model path
   print_interval: 1
 - name: infer_runner
-  class: single_infer
-  # num of epochs
-  epochs: 1
+  class: infer
   # device to run training or infer
   device: cpu
   print_interval: 1

models/match/readme.md

@@ -43,7 +43,7 @@ python -m paddlerec.run -m paddlerec.models.match.multiview-simnet # multiview-s
 # 修改对应模型的config.yaml, workspace配置为当前目录的绝对路径
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: train_runner -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 修改phase阶段为infer的配置，参照config注释
 
 # 修改完config.yaml后 执行:

models/multitask/esmm/config.yaml

@@ -16,21 +16,21 @@
 workspace: "paddlerec.models.multitask.esmm"
 
 dataset:
-- name: dataset_train
-  batch_size: 1
-  type: QueueDataset
-  data_path: "{workspace}/data/train"
-  data_converter: "{workspace}/esmm_reader.py"
-- name: dataset_infer
-  batch_size: 1
-  type: QueueDataset
-  data_path: "{workspace}/data/test"
-  data_converter: "{workspace}/esmm_reader.py"
+  - name: dataset_train
+    batch_size: 1
+    type: QueueDataset
+    data_path: "{workspace}/data/train"
+    data_converter: "{workspace}/esmm_reader.py"
+  - name: dataset_infer
+    batch_size: 1
+    type: QueueDataset
+    data_path: "{workspace}/data/test"
+    data_converter: "{workspace}/esmm_reader.py"
 
 hyper_parameters:
   vocab_size: 10000
   embed_size: 128
-  optimizer: 
+  optimizer:
     class: adam
     learning_rate: 0.001
     strategy: async
@@ -39,30 +39,29 @@ hyper_parameters:
 mode: [train_runner, infer_runner]
 
 runner:
-- name: train_runner
-  class: train
-  device: cpu
-  epochs: 3
-  save_checkpoint_interval: 2
-  save_inference_interval: 4
-  save_checkpoint_path: "increment"
-  save_inference_path: "inference"
-  print_interval: 10
-  phases: [train]
-- name: infer_runner
-  class: infer
-  init_model_path: "increment/0"
-  device: cpu
-  epochs: 1
-  print_interval: 1
-  phases: [infer]
+  - name: train_runner
+    class: train
+    device: cpu
+    epochs: 3
+    save_checkpoint_interval: 2
+    save_inference_interval: 4
+    save_checkpoint_path: "increment"
+    save_inference_path: "inference"
+    print_interval: 10
+    phases: [train]
+  - name: infer_runner
+    class: infer
+    init_model_path: "increment/1"
+    device: cpu
+    print_interval: 1
+    phases: [infer]
 
 phase:
-- name: train
-  model: "{workspace}/model.py"
-  dataset_name: dataset_train
-  thread_num: 1
-- name: infer
-  model: "{workspace}/model.py"
-  dataset_name: dataset_infer
-  thread_num: 1
+  - name: train
+    model: "{workspace}/model.py"
+    dataset_name: dataset_train
+    thread_num: 1
+  - name: infer
+    model: "{workspace}/model.py"
+    dataset_name: dataset_infer
+    thread_num: 1

models/multitask/mmoe/config.yaml

@@ -52,10 +52,9 @@ runner:
   save_inference_path: "inference"
   print_interval: 10
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 3
 
 phase:
 - name: train

models/multitask/readme.md

@@ -77,7 +77,7 @@ python -m paddlerec.run -m ./config.yaml # 自定义修改超参后，指定配
 # 修改对应模型的config.yaml, workspace配置为当前目录的绝对路径
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: train_runner -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 修改phase阶段为infer的配置，参照config注释
 
 # 修改完config.yaml后 执行:

models/multitask/share-bottom/config.yaml

@@ -51,10 +51,9 @@ runner:
   save_inference_path: "inference"
   print_interval: 5
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 3
 
 phase:
 - name: train

models/rank/afm/config.yaml

@@ -59,8 +59,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/dcn/config.yaml

@@ -60,7 +60,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
+    class: infer
     epochs: 1
     device: cpu
     init_model_path: "increment/0"

models/rank/deep_crossing/config.yaml

@@ -58,8 +58,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/deepfm/config.yaml

@@ -58,8 +58,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/din/config.yaml

@@ -56,8 +56,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/ffm/config.yaml

@@ -57,8 +57,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/fgcnn/config.yaml

@@ -62,8 +62,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/fm/config.yaml

@@ -57,8 +57,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/fnn/config.yaml

@@ -68,8 +68,7 @@ runner:
     save_inference_path: "inference_fnn"
     print_interval: 1
   - name: infer_runner
-    trainer_class: single_infer
-    epochs: 1
+    trainer_class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/logistic_regression/config.yaml

@@ -56,8 +56,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/nfm/config.yaml

@@ -63,8 +63,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/pnn/config.yaml

@@ -60,8 +60,7 @@ runner:
     save_inference_path: "inference"
     print_interval: 1
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
     print_interval: 1

models/rank/readme.md

@@ -98,7 +98,7 @@ python -m paddlerec.run -m ./config.yaml
 ```
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: runner1 -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 如果训练阶段和预测阶段的模型输入一致，phase不需要改动，复用train的即可
 
 # 修改完config.yaml后 执行:

models/rank/wide_deep/config.yaml

@@ -54,8 +54,7 @@ runner:
     save_checkpoint_path: "increment"
     save_inference_path: "inference"
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
 

models/rank/xdeepfm/config.yaml

@@ -55,8 +55,7 @@ runner:
     save_checkpoint_path: "increment"
     save_inference_path: "inference"
   - name: infer_runner
-    class: single_infer
-    epochs: 1
+    class: infer
     device: cpu
     init_model_path: "increment/0"
 

models/recall/gnn/config.yaml

@@ -61,9 +61,7 @@ runner:
   init_model_path: "" # load model path
   print_interval: 1
 - name: infer_runner
-  class: single_infer
-  # num of epochs
-  epochs: 1
+  class: infer
   # device to run training or infer
   device: cpu
   print_interval: 1

models/recall/gru4rec/config.yaml

@@ -54,10 +54,9 @@ runner:
   save_inference_path: "inference"
   print_interval: 10
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 1
 
 phase:
 - name: train

models/recall/ncf/config.yaml

@@ -51,10 +51,9 @@ runner:
   save_inference_path: "inference"
   print_interval: 10
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 1
 
 phase:
 - name: train

models/recall/readme.md

@@ -95,7 +95,7 @@ python -m paddlerec.run -m ./config.yaml # 自定义修改超参后，指定配
 # 修改对应模型的config.yaml, workspace配置为当前目录的绝对路径
 # 修改对应模型的config.yaml，mode配置infer_runner
 # 示例: mode: train_runner -> mode: infer_runner
-# infer_runner中 class配置为 class: single_infer
+# infer_runner中 class配置为 class: infer
 # 修改phase阶段为infer的配置，参照config注释
 
 # 修改完config.yaml后 执行:

models/recall/ssr/config.yaml

@@ -50,10 +50,9 @@ runner:
   save_inference_path: "inference"
   print_interval: 10
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 1
 
 phase:
 - name: train

models/recall/word2vec/config.yaml

@@ -61,9 +61,7 @@ runner:
   init_model_path: "" # load model path
   print_interval: 1
 - name: infer_runner
-  class: single_infer
-  # num of epochs
-  epochs: 1
+  class: infer
   # device to run training or infer
   device: cpu
   init_model_path: "increment/0" # load model path

models/rerank/listwise/config.yaml

@@ -51,10 +51,9 @@ runner:
   save_checkpoint_path: "increment"
   save_inference_path: "inference"
 - name: infer_runner
-  class: single_infer
+  class: infer
   init_model_path: "increment/0"
   device: cpu
-  epochs: 3
 
 phase:
 - name: train

models/treebased/tdm/config.yaml

@@ -80,10 +80,8 @@ runner:
   print_interval: 10
 
 - name: runner2
-  class: single_infer
+  class: infer
   startup_class_path: "{workspace}/tdm_startup.py"
-  # num of epochs
-  epochs: 1
   # device to run training or infer
   device: cpu
   init_model_path: "increment/0" # load model path

run.py

@@ -139,8 +139,8 @@ def get_engine(args, running_config, mode):
             engine = "LOCAL_CLUSTER_TRAIN"
 
     if engine not in engine_choices:
-        raise ValueError("{} can not be chosen in {}".format(engine_class,
-                                                             engine_choices))
+        raise ValueError("{} can only be chosen in {}".format(engine_class,
+                                                              engine_choices))
 
     run_engine = engines[transpiler].get(engine, None)
     return run_engine
@@ -439,8 +439,8 @@ def local_cluster_engine(args):
     if fleet_mode == "COLLECTIVE":
         cluster_envs["selected_gpus"] = selected_gpus
         gpus = selected_gpus.split(",")
-        gpu_num = get_worker_num(run_extras, len(gpus))
-        cluster_envs["selected_gpus"] = ','.join(gpus[:gpu_num])
+        worker_num = get_worker_num(run_extras, len(gpus))
+        cluster_envs["selected_gpus"] = ','.join(gpus[:worker_num])
 
     cluster_envs["server_num"] = server_num
     cluster_envs["worker_num"] = worker_num

tools/build_script.sh

@@ -49,7 +49,7 @@ function model_test() {
   
   root_dir=`pwd`
   all_model=$(find ${root_dir} -name config.yaml)
-  special_models=("demo" "pnn" "fgcnn" "esmm")
+  special_models=("demo" "pnn" "fgcnn" "gru4rec" "tagspace")
 
   for model in ${all_model}
   do