test_downpoursgd.py 8.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
#   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.
14
"""Testcases for Downpour."""
15 16 17 18 19 20 21 22 23 24 25 26 27

from __future__ import print_function

import paddle
import paddle.fluid as fluid
import os
import signal
import subprocess
import time
import unittest
import sys
from op_test import OpTest
from paddle.fluid.trainer_desc import DistMultiTrainer
28
from paddle.fluid.device_worker import DownpourSGD, DownpourSGDOPT
29
from paddle.fluid.incubate.fleet.parameter_server.pslib.node import DownpourWorker
30 31
from google.protobuf import text_format
import paddle.fluid.incubate.fleet.parameter_server.pslib.ps_pb2 as pslib
32
from paddle.fluid.trainer_factory import TrainerFactory
33 34


35
class TestListenAndServOp(unittest.TestCase):
36 37
    """TestListenAndServOp."""

38 39 40 41
    def setUp(self):
        pass

    def test_device_work_use_cvm(self):
42
        """test device work use_cvm."""
43 44 45 46 47 48
        if sys.platform == 'win32' or sys.platform == 'sys.platform':
            pass
        else:
            print(sys.platform)
            cmd = "wget --no-check-certificate https://pslib.bj.bcebos.com/fleet_desc.prototxt"
            os.system(cmd)
49
            x = fluid.layers.data(name='x', shape=[1], dtype='int64')
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
            x_emb = fluid.layers.embedding(
                input=x, size=[1, 2], is_distributed=True)
            y_predict = fluid.layers.fc(input=x_emb, size=1, act=None)
            y = fluid.layers.data(name='y', shape=[1], dtype='float32')
            cost = fluid.layers.square_error_cost(input=y_predict, label=y)
            avg_cost = fluid.layers.mean(cost)

            ps_param = pslib.PSParameter()
            with open("fleet_desc.prototxt") as f:
                text_format.Merge(f.read(), ps_param)
            fleet_desc = ps_param
            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())

            opt_info = {}
            main_program = fluid.default_main_program()
            program_id = str(id(avg_cost.block.program))
            program_configs = {}
            program_configs[program_id] = {
                "pull_sparse": [0],
                "push_sparse": [0]
            }
            program_configs[program_id]["pull_dense"] = [1]
            program_configs[program_id]["push_dense"] = [1]

            worker_skipped_ops = ["lookup_table", "lookup_table_grad"]
            opt_info["program_configs"] = program_configs
            opt_info["trainer"] = "DistMultiTrainer"
            opt_info["device_worker"] = "DownpourSGD"
            opt_info["optimizer"] = "DownpourSGD"
            opt_info["fleet_desc"] = ps_param
            opt_info["worker_skipped_ops"] = worker_skipped_ops
            opt_info["use_cvm"] = True
            opt_info["scale_datanorm"] = -1
            opt_info["dump_slot"] = False
85
            opt_info["stat_var_names"] = []
86 87 88
            worker = DownpourWorker(None)
            worker.get_desc().CopyFrom(ps_param.trainer_param[0])
            opt_info["program_id_to_worker"] = {program_id: worker}
89 90

            main_program._fleet_opt = opt_info
91
            trainer = TrainerFactory()._create_trainer(main_program._fleet_opt)
92 93 94 95 96 97
            trainer._set_program(main_program)
            trainer._gen_trainer_desc()
            cmd = "rm fleet_desc.prototxt*"
            os.system(cmd)

    def test_device_work(self):
98
        """test devicve worker."""
99 100 101 102 103 104
        if sys.platform == 'win32' or sys.platform == 'sys.platform':
            pass
        else:
            print(sys.platform)
            cmd = "wget --no-check-certificate https://pslib.bj.bcebos.com/fleet_desc.prototxt"
            os.system(cmd)
105
            x = fluid.layers.data(name='x', shape=[1], dtype='int64')
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
            x_emb = fluid.layers.embedding(
                input=x, size=[1, 2], is_distributed=True)
            y_predict = fluid.layers.fc(input=x_emb, size=1, act=None)
            y = fluid.layers.data(name='y', shape=[1], dtype='float32')
            cost = fluid.layers.square_error_cost(input=y_predict, label=y)
            avg_cost = fluid.layers.mean(cost)

            ps_param = pslib.PSParameter()
            with open("fleet_desc.prototxt") as f:
                text_format.Merge(f.read(), ps_param)
            fleet_desc = ps_param
            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())

            opt_info = {}
            main_program = fluid.default_main_program()
            program_id = str(id(avg_cost.block.program))
            program_configs = {}
            program_configs[program_id] = {
                "pull_sparse": [0],
                "push_sparse": [0]
            }
            program_configs[program_id]["pull_dense"] = [1]
            program_configs[program_id]["push_dense"] = [1]

            worker_skipped_ops = ["lookup_table", "lookup_table_grad"]
            opt_info["program_configs"] = program_configs
            opt_info["trainer"] = "DistMultiTrainer"
            opt_info["device_worker"] = "DownpourSGD"
            opt_info["optimizer"] = "DownpourSGD"
            opt_info["fleet_desc"] = ps_param
            opt_info["worker_skipped_ops"] = worker_skipped_ops
            opt_info["use_cvm"] = False
            opt_info["scale_datanorm"] = -1
            opt_info["dump_slot"] = False
141
            opt_info["stat_var_names"] = []
142 143 144
            worker = DownpourWorker(None)
            worker.get_desc().CopyFrom(ps_param.trainer_param[0])
            opt_info["program_id_to_worker"] = {program_id: worker}
145 146

            main_program._fleet_opt = opt_info
147
            trainer = TrainerFactory()._create_trainer(main_program._fleet_opt)
148 149 150 151 152
            trainer._set_program(main_program)
            trainer._gen_trainer_desc()
            cmd = "rm fleet_desc.prototxt*"
            os.system(cmd)

153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
    def test_downpour_opt_work(self):
        """test devicve worker."""
        if sys.platform == 'win32' or sys.platform == 'sys.platform':
            pass
        else:
            print(sys.platform)
            cmd = "wget --no-check-certificate https://pslib.bj.bcebos.com/fleet_desc.prototxt"
            os.system(cmd)
            x = fluid.layers.data(name='x', shape=[1], dtype='int64')
            x_emb = fluid.layers.embedding(
                input=x, size=[1, 2], is_distributed=True)
            y_predict = fluid.layers.fc(input=x_emb, size=1, act=None)
            y = fluid.layers.data(name='y', shape=[1], dtype='float32')
            cost = fluid.layers.square_error_cost(input=y_predict, label=y)
            avg_cost = fluid.layers.mean(cost)

            ps_param = pslib.PSParameter()
            with open("fleet_desc.prototxt") as f:
                text_format.Merge(f.read(), ps_param)
            fleet_desc = ps_param
            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())

            opt_info = {}
            main_program = fluid.default_main_program()
            program_id = str(id(avg_cost.block.program))
            program_configs = {}
            program_configs[program_id] = {
                "pull_sparse": [0],
                "push_sparse": [0]
            }
            program_configs[program_id]["pull_dense"] = [1]
            program_configs[program_id]["push_dense"] = [1]

            worker_skipped_ops = ["lookup_table", "lookup_table_grad"]
            opt_info["program_configs"] = program_configs
            opt_info["trainer"] = "DistMultiTrainer"
            opt_info["device_worker"] = "DownpourSGDOPT"
            opt_info["optimizer"] = "DownpourSGD"
            opt_info["fleet_desc"] = ps_param
            opt_info["worker_skipped_ops"] = worker_skipped_ops
            opt_info["use_cvm"] = False
            opt_info["scale_datanorm"] = -1
            opt_info["dump_slot"] = False
            opt_info["stat_var_names"] = []
            worker = DownpourWorker(None)
            worker.get_desc().CopyFrom(ps_param.trainer_param[0])
            opt_info["program_id_to_worker"] = {program_id: worker}

            main_program._fleet_opt = opt_info
203
            trainer = TrainerFactory()._create_trainer(main_program._fleet_opt)
204 205 206 207 208
            trainer._set_program(main_program)
            trainer._gen_trainer_desc()
            cmd = "rm fleet_desc.prototxt*"
            os.system(cmd)

209 210 211

if __name__ == "__main__":
    unittest.main()