support windows system (#215)

* first version; unittest in parl/remote is passed

* add windows local unittest shell; fix some incompatible problem

* fix save api

* refine comments and remove log of xparl stop

* fix bug

* Update scripts.py

* Update utils.py

* Update agent_base_test.py

Author: Hongsheng Zeng

.teamcity/requirements.txt

@@ -3,4 +3,3 @@ paddlepaddle-gpu==1.6.1.post97
 gym
 details
 parameterized
-timeout_decorator

.teamcity/requirements_torch.txt

@@ -2,4 +2,3 @@
 gym
 details
 parameterized
-timeout_decorator

.teamcity/windows_test.sh

@@ -0,0 +1,76 @@
+#!/usr/bin/env bash
+
+# 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.
+
+# NOTE: You need install mingw-cmake.
+
+function init() {
+    RED='\033[0;31m'
+    BLUE='\033[0;34m'
+    BOLD='\033[1m'
+    NONE='\033[0m'
+
+    REPO_ROOT=`pwd`
+}
+
+
+function abort(){
+    echo "Your change doesn't follow PaddlePaddle's code style." 1>&2
+    echo "Please use pre-commit to check what is wrong." 1>&2
+    exit 1
+}
+
+function run_test_with_cpu() {
+    export CUDA_VISIBLE_DEVICES="-1"
+
+    mkdir -p ${REPO_ROOT}/build
+    cd ${REPO_ROOT}/build
+    if [ $# -eq 1 ];then
+        cmake  -G "MinGW Makefiles" ..
+    else
+        cmake  -G "MinGW Makefiles" .. -$2=ON
+    fi
+    cat <<EOF
+    =====================================================
+    Running unit tests with CPU in the environment: $1
+    =====================================================
+EOF
+    if [ $# -eq 1 ];then
+      ctest --output-on-failure -j10
+    else
+      ctest --output-on-failure 
+    fi
+    cd ${REPO_ROOT}
+    rm -rf ${REPO_ROOT}/build
+}
+
+function main() {
+    set -e
+    local CMD=$1
+    
+    init
+    env="unused_variable"
+    # run unittest in windows (used in local machine)
+    pip install -i https://pypi.tuna.tsinghua.edu.cn/simple .
+    pip uninstall -y torch torchvision
+    pip install -i https://pypi.tuna.tsinghua.edu.cn/simple paddlepaddle==1.6.1 gym details parameterized
+    run_test_with_cpu $env
+    run_test_with_cpu $env "DIS_TESTING_SERIALLY"
+    pip uninstall -y paddlepaddle
+    pip install torch==1.4.0+cpu torchvision==0.5.0+cpu -f https://download.pytorch.org/whl/torch_stable.html  
+    run_test_with_cpu $env "DIS_TESTING_TORCH"
+}
+
+main $@

parl/core/fluid/agent.py

@@ -15,6 +15,7 @@
 import warnings
 warnings.simplefilter('default')
 
+import os
 import paddle.fluid as fluid
 from parl.core.fluid import layers
 from parl.core.agent_base import AgentBase
@@ -152,8 +153,8 @@ def save(self, save_path, program=None):
         """
         if program is None:
             program = self.learn_program
-        dirname = '/'.join(save_path.split('/')[:-1])
-        filename = save_path.split('/')[-1]
+        dirname = os.sep.join(save_path.split(os.sep)[:-1])
+        filename = save_path.split(os.sep)[-1]
         fluid.io.save_params(
             executor=self.fluid_executor,
             dirname=dirname,
@@ -186,8 +187,8 @@ def restore(self, save_path, program=None):
             program = self.learn_program
         if type(program) is fluid.compiler.CompiledProgram:
             program = program._init_program
-        dirname = '/'.join(save_path.split('/')[:-1])
-        filename = save_path.split('/')[-1]
+        dirname = os.sep.join(save_path.split(os.sep)[:-1])
+        filename = save_path.split(os.sep)[-1]
         fluid.io.load_params(
             executor=self.fluid_executor,
             dirname=dirname,

parl/core/fluid/layers/tests/param_sharing_test.py

@@ -45,7 +45,7 @@ def test_param_sharing(self):
         dict_size = 100
         input_cx = np.random.uniform(0, 1, [batch_size, 100]).astype("float32")
         input_x = np.random.randint(
-            dict_size, size=(batch_size, 1)).astype("int")
+            dict_size, size=(batch_size, 1)).astype("int64")
         #################################
 
         main_program1 = fluid.Program()
@@ -59,7 +59,7 @@ def test_param_sharing(self):
 
         main_program2 = fluid.Program()
         with fluid.program_guard(main_program2):
-            x_ = layers.data(name='x', shape=[1], dtype="int")
+            x_ = layers.data(name='x', shape=[1], dtype="int64")
             cx_ = layers.cast(
                 x=layers.one_hot(input=x_, depth=dict_size), dtype="float32")
             y1_ = net.fc1(input=cx_)

parl/core/fluid/tests/agent_base_test_.py renamed to parl/core/fluid/tests/agent_base_test.py

@@ -46,8 +46,8 @@ def learn(self, obs, label):
 
 
 class TestAgent(parl.Agent):
-    def __init__(self, algorithm, gpu_id=None):
-        super(TestAgent, self).__init__(algorithm, gpu_id)
+    def __init__(self, algorithm):
+        super(TestAgent, self).__init__(algorithm)
 
     def build_program(self):
         self.predict_program = fluid.Program()
@@ -92,8 +92,8 @@ def test_save(self):
         agent = TestAgent(self.algorithm)
         obs = np.random.random([3, 10]).astype('float32')
         output_np = agent.predict(obs)
-        save_path1 = './model.ckpt'
-        save_path2 = './my_model/model-2.ckpt'
+        save_path1 = 'model.ckpt'
+        save_path2 = os.path.join('my_model', 'model-2.ckpt')
         agent.save(save_path1)
         agent.save(save_path2)
         self.assertTrue(os.path.exists(save_path1))
@@ -103,7 +103,7 @@ def test_restore(self):
         agent = TestAgent(self.algorithm)
         obs = np.random.random([3, 10]).astype('float32')
         output_np = agent.predict(obs)
-        save_path1 = './model.ckpt'
+        save_path1 = 'model.ckpt'
         previous_output = agent.predict(obs)
         agent.save(save_path1)
         agent.restore(save_path1)
@@ -121,7 +121,7 @@ def test_compiled_restore(self):
         agent.learn_program = parl.compile(agent.learn_program)
         obs = np.random.random([3, 10]).astype('float32')
         previous_output = agent.predict(obs)
-        save_path1 = './model.ckpt'
+        save_path1 = 'model.ckpt'
         agent.save(save_path1)
         agent.restore(save_path1)
 

parl/core/fluid/tests/model_base_test_.py renamed to parl/core/fluid/tests/model_base_test.py

@@ -113,8 +113,9 @@ def save(self, save_path, model=None):
         """
         if model is None:
             model = self.algorithm.model
-        dirname = '/'.join(save_path.split('/')[:-1])
-        if not os.path.exists(dirname):
+        sep = os.sep
+        dirname = sep.join(save_path.split(sep)[:-1])
+        if dirname != '' and not os.path.exists(dirname):
             os.makedirs(dirname)
         torch.save(model.state_dict(), save_path)
 

parl/core/fluid/tests/policy_distribution_test_.py renamed to parl/core/fluid/tests/policy_distribution_test.py

@@ -77,8 +77,8 @@ def test_agent(self):
     def test_save(self):
         agent = TestAgent(self.alg)
         obs = torch.randn(3, 10)
-        save_path1 = './model.ckpt'
-        save_path2 = './my_model/model-2.ckpt'
+        save_path1 = 'model.ckpt'
+        save_path2 = os.path.join('my_model', 'model-2.ckpt')
         agent.save(save_path1)
         agent.save(save_path2)
         self.assertTrue(os.path.exists(save_path1))
@@ -88,7 +88,7 @@ def test_restore(self):
         agent = TestAgent(self.alg)
         obs = torch.randn(3, 10)
         output = agent.predict(obs)
-        save_path1 = './model.ckpt'
+        save_path1 = 'model.ckpt'
         previous_output = agent.predict(obs).detach().cpu().numpy()
         agent.save(save_path1)
         agent.restore(save_path1)