Add C++ inference unittest of recommender system (#8227)

Xreki · kexinzhao · commit 1185a1b5ab89 · 2018-02-08T21:27:42.000-08:00
* Save the inference model in Python example of recommender_system.

* Add infer() in Python unittest recommender_system.

* Add C++ inference unittest of recommender_system.
diff --git a/paddle/inference/tests/book/CMakeLists.txt b/paddle/inference/tests/book/CMakeLists.txt
@@ -28,3 +28,4 @@ inference_test(recognize_digits ARGS mlp)
 inference_test(image_classification ARGS vgg resnet)
 inference_test(label_semantic_roles)
 inference_test(rnn_encoder_decoder)
+inference_test(recommender_system)
diff --git a/paddle/inference/tests/book/test_helper.h b/paddle/inference/tests/book/test_helper.h
@@ -30,14 +30,34 @@ void SetupTensor(paddle::framework::LoDTensor& input,
   }
 }
 
+template <typename T>
+void SetupTensor(paddle::framework::LoDTensor& input,
+                 paddle::framework::DDim dims,
+                 std::vector<T>& data) {
+  CHECK_EQ(paddle::framework::product(dims), data.size());
+  T* input_ptr = input.mutable_data<T>(dims, paddle::platform::CPUPlace());
+  memcpy(input_ptr, data.data(), input.numel() * sizeof(T));
+}
+
 template <typename T>
 void SetupLoDTensor(paddle::framework::LoDTensor& input,
                     paddle::framework::LoD& lod,
                     T lower,
                     T upper) {
   input.set_lod(lod);
   int dim = lod[0][lod[0].size() - 1];
-  SetupTensor(input, {dim, 1}, lower, upper);
+  SetupTensor<T>(input, {dim, 1}, lower, upper);
+}
+
+template <typename T>
+void SetupLoDTensor(paddle::framework::LoDTensor& input,
+                    paddle::framework::DDim dims,
+                    paddle::framework::LoD lod,
+                    std::vector<T>& data) {
+  const size_t level = lod.size() - 1;
+  CHECK_EQ(dims[0], (lod[level]).back());
+  input.set_lod(lod);
+  SetupTensor<T>(input, dims, data);
 }
 
 template <typename T>
@@ -67,26 +87,29 @@ void CheckError(paddle::framework::LoDTensor& output1,
   EXPECT_EQ(count, 0) << "There are " << count << " different elements.";
 }
 
-template <typename Place, typename T, bool IsCombined = false>
+template <typename Place, bool IsCombined = false>
 void TestInference(const std::string& dirname,
                    const std::vector<paddle::framework::LoDTensor*>& cpu_feeds,
                    std::vector<paddle::framework::LoDTensor*>& cpu_fetchs) {
   // 1. Define place, executor, scope and inference_program
   auto place = Place();
   auto executor = paddle::framework::Executor(place);
   auto* scope = new paddle::framework::Scope();
-  std::unique_ptr<paddle::framework::ProgramDesc> inference_program;
 
-  // 2. Initialize the inference_program and load all parameters from file
+  // 2. Initialize the inference_program and load parameters
+  std::unique_ptr<paddle::framework::ProgramDesc> inference_program;
   if (IsCombined) {
-    // Hard-coding the names for combined params case
+    // All parameters are saved in a single file.
+    // Hard-coding the file names of program and parameters in unittest.
+    // Users are free to specify different filename.
     std::string prog_filename = "__model_combined__";
     std::string param_filename = "__params_combined__";
     inference_program = paddle::inference::Load(executor,
                                                 *scope,
                                                 dirname + "/" + prog_filename,
                                                 dirname + "/" + param_filename);
   } else {
+    // Parameters are saved in separate files sited in the specified `dirname`.
     inference_program = paddle::inference::Load(executor, *scope, dirname);
   }
 
diff --git a/paddle/inference/tests/book/test_inference_image_classification.cc b/paddle/inference/tests/book/test_inference_image_classification.cc
@@ -29,11 +29,15 @@ TEST(inference, image_classification) {
   // 0. Call `paddle::framework::InitDevices()` initialize all the devices
   // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
 
+  int64_t batch_size = 1;
+
   paddle::framework::LoDTensor input;
   // Use normilized image pixels as input data,
   // which should be in the range [0.0, 1.0].
-  SetupTensor<float>(
-      input, {1, 3, 32, 32}, static_cast<float>(0), static_cast<float>(1));
+  SetupTensor<float>(input,
+                     {batch_size, 3, 32, 32},
+                     static_cast<float>(0),
+                     static_cast<float>(1));
   std::vector<paddle::framework::LoDTensor*> cpu_feeds;
   cpu_feeds.push_back(&input);
 
@@ -42,8 +46,7 @@ TEST(inference, image_classification) {
   cpu_fetchs1.push_back(&output1);
 
   // Run inference on CPU
-  TestInference<paddle::platform::CPUPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs1);
+  TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
   LOG(INFO) << output1.dims();
 
 #ifdef PADDLE_WITH_CUDA
@@ -52,8 +55,7 @@ TEST(inference, image_classification) {
   cpu_fetchs2.push_back(&output2);
 
   // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs2);
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.dims();
 
   CheckError<float>(output1, output2);
diff --git a/paddle/inference/tests/book/test_inference_label_semantic_roles.cc b/paddle/inference/tests/book/test_inference_label_semantic_roles.cc
@@ -58,8 +58,7 @@ TEST(inference, label_semantic_roles) {
   cpu_fetchs1.push_back(&output1);
 
   // Run inference on CPU
-  TestInference<paddle::platform::CPUPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs1);
+  TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
   LOG(INFO) << output1.lod();
   LOG(INFO) << output1.dims();
 
@@ -69,8 +68,7 @@ TEST(inference, label_semantic_roles) {
   cpu_fetchs2.push_back(&output2);
 
   // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs2);
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.lod();
   LOG(INFO) << output2.dims();
 
diff --git a/paddle/inference/tests/book/test_inference_recognize_digits.cc b/paddle/inference/tests/book/test_inference_recognize_digits.cc
@@ -29,11 +29,15 @@ TEST(inference, recognize_digits) {
   // 0. Call `paddle::framework::InitDevices()` initialize all the devices
   // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
 
+  int64_t batch_size = 1;
+
   paddle::framework::LoDTensor input;
   // Use normilized image pixels as input data,
   // which should be in the range [-1.0, 1.0].
-  SetupTensor<float>(
-      input, {1, 28, 28}, static_cast<float>(-1), static_cast<float>(1));
+  SetupTensor<float>(input,
+                     {batch_size, 1, 28, 28},
+                     static_cast<float>(-1),
+                     static_cast<float>(1));
   std::vector<paddle::framework::LoDTensor*> cpu_feeds;
   cpu_feeds.push_back(&input);
 
@@ -42,8 +46,7 @@ TEST(inference, recognize_digits) {
   cpu_fetchs1.push_back(&output1);
 
   // Run inference on CPU
-  TestInference<paddle::platform::CPUPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs1);
+  TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
   LOG(INFO) << output1.dims();
 
 #ifdef PADDLE_WITH_CUDA
@@ -52,8 +55,7 @@ TEST(inference, recognize_digits) {
   cpu_fetchs2.push_back(&output2);
 
   // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs2);
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.dims();
 
   CheckError<float>(output1, output2);
@@ -84,7 +86,7 @@ TEST(inference, recognize_digits_combine) {
   cpu_fetchs1.push_back(&output1);
 
   // Run inference on CPU
-  TestInference<paddle::platform::CPUPlace, float, true>(
+  TestInference<paddle::platform::CPUPlace, true>(
       dirname, cpu_feeds, cpu_fetchs1);
   LOG(INFO) << output1.dims();
 
@@ -94,7 +96,7 @@ TEST(inference, recognize_digits_combine) {
   cpu_fetchs2.push_back(&output2);
 
   // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace, float, true>(
+  TestInference<paddle::platform::CUDAPlace, true>(
       dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.dims();
 
diff --git a/paddle/inference/tests/book/test_inference_recommender_system.cc b/paddle/inference/tests/book/test_inference_recommender_system.cc
@@ -0,0 +1,87 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+
+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. */
+
+#include <gtest/gtest.h>
+#include "gflags/gflags.h"
+#include "test_helper.h"
+
+DEFINE_string(dirname, "", "Directory of the inference model.");
+
+TEST(inference, recommender_system) {
+  if (FLAGS_dirname.empty()) {
+    LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
+  }
+
+  LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
+  std::string dirname = FLAGS_dirname;
+
+  // 0. Call `paddle::framework::InitDevices()` initialize all the devices
+  // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
+
+  int64_t batch_size = 1;
+
+  paddle::framework::LoDTensor user_id, gender_id, age_id, job_id, movie_id,
+      category_id, movie_title;
+
+  // Use the first data from paddle.dataset.movielens.test() as input
+  std::vector<int64_t> user_id_data = {1};
+  SetupTensor<int64_t>(user_id, {batch_size, 1}, user_id_data);
+
+  std::vector<int64_t> gender_id_data = {1};
+  SetupTensor<int64_t>(gender_id, {batch_size, 1}, gender_id_data);
+
+  std::vector<int64_t> age_id_data = {0};
+  SetupTensor<int64_t>(age_id, {batch_size, 1}, age_id_data);
+
+  std::vector<int64_t> job_id_data = {10};
+  SetupTensor<int64_t>(job_id, {batch_size, 1}, job_id_data);
+
+  std::vector<int64_t> movie_id_data = {783};
+  SetupTensor<int64_t>(movie_id, {batch_size, 1}, movie_id_data);
+
+  std::vector<int64_t> category_id_data = {10, 8, 9};
+  SetupLoDTensor<int64_t>(category_id, {3, 1}, {{0, 3}}, category_id_data);
+
+  std::vector<int64_t> movie_title_data = {1069, 4140, 2923, 710, 988};
+  SetupLoDTensor<int64_t>(movie_title, {5, 1}, {{0, 5}}, movie_title_data);
+
+  std::vector<paddle::framework::LoDTensor*> cpu_feeds;
+  cpu_feeds.push_back(&user_id);
+  cpu_feeds.push_back(&gender_id);
+  cpu_feeds.push_back(&age_id);
+  cpu_feeds.push_back(&job_id);
+  cpu_feeds.push_back(&movie_id);
+  cpu_feeds.push_back(&category_id);
+  cpu_feeds.push_back(&movie_title);
+
+  paddle::framework::LoDTensor output1;
+  std::vector<paddle::framework::LoDTensor*> cpu_fetchs1;
+  cpu_fetchs1.push_back(&output1);
+
+  // Run inference on CPU
+  TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
+  LOG(INFO) << output1.dims();
+
+#ifdef PADDLE_WITH_CUDA
+  paddle::framework::LoDTensor output2;
+  std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
+  cpu_fetchs2.push_back(&output2);
+
+  // Run inference on CUDA GPU
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
+  LOG(INFO) << output2.dims();
+
+  CheckError<float>(output1, output2);
+#endif
+}
diff --git a/paddle/inference/tests/book/test_inference_rnn_encoder_decoder.cc b/paddle/inference/tests/book/test_inference_rnn_encoder_decoder.cc
@@ -46,8 +46,7 @@ TEST(inference, rnn_encoder_decoder) {
   cpu_fetchs1.push_back(&output1);
 
   // Run inference on CPU
-  TestInference<paddle::platform::CPUPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs1);
+  TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
   LOG(INFO) << output1.lod();
   LOG(INFO) << output1.dims();
 
@@ -57,8 +56,7 @@ TEST(inference, rnn_encoder_decoder) {
   cpu_fetchs2.push_back(&output2);
 
   // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace, float>(
-      dirname, cpu_feeds, cpu_fetchs2);
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.lod();
   LOG(INFO) << output2.dims();
 
diff --git a/python/paddle/v2/fluid/tests/book/.gitignore b/python/paddle/v2/fluid/tests/book/.gitignore
@@ -1 +1 @@
-recognize_digits_*.inference.model
+*.inference.model
diff --git a/python/paddle/v2/fluid/tests/book/test_recognize_digits.py b/python/paddle/v2/fluid/tests/book/test_recognize_digits.py
@@ -174,8 +174,9 @@ def infer(use_cuda, save_dirname=None, param_filename=None):
 
     # The input's dimension of conv should be 4-D or 5-D.
     # Use normilized image pixels as input data, which should be in the range [-1.0, 1.0].
+    batch_size = 1
     tensor_img = numpy.random.uniform(-1.0, 1.0,
-                                      [1, 1, 28, 28]).astype("float32")
+                                      [batch_size, 1, 28, 28]).astype("float32")
 
     # Construct feed as a dictionary of {feed_target_name: feed_target_data}
     # and results will contain a list of data corresponding to fetch_targets.
diff --git a/python/paddle/v2/fluid/tests/book/test_recommender_system.py b/python/paddle/v2/fluid/tests/book/test_recommender_system.py

Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-recognize_digits_*.inference.model`
	`1`	`+*.inference.model`