Machine translation example

src/example/CMakeLists.txt

@@ -15,3 +15,4 @@
 
 add_subdirectory(mnist)
 add_subdirectory(cifar)
+add_subdirectory(machine_translation)

src/example/cifar/test_cifar.cc

@@ -50,10 +50,10 @@ using paddle::tape::ResetReader;
 
 TEST(Cifar, TestGPU) {
   // auto place = paddle::platform::CPUPlace();
-  auto place = paddle::platform::CUDAPlace(0);
+  auto place = paddle::platform::CUDAPlace(1);
   reset_global_tape(place);
 
-  const int batch_size = 128;
+  const int batch_size = 64;
   LOG(INFO) << "Batch size is " << batch_size << std::endl;
 
   std::string save_model_path = "/tmp/cifar_model/";
@@ -119,10 +119,10 @@ TEST(Cifar, TestGPU) {
   BatchNorm bn5_3(512, "relu");
 
   // Input dim 512x1x1 = 512
-  Linear fc1(512, 512);
+  Linear fc1({512}, 512);
   BatchNorm bn6(512, "relu");
-  Linear fc2(512, 512);
-  Linear fc3(512, 10, "softmax");
+  Linear fc2({512}, 512);
+  Linear fc3({512}, 10, "softmax");
 
   Adam adam(0.001);
 
@@ -153,19 +153,19 @@ TEST(Cifar, TestGPU) {
     auto pool5 = pool2d(bn5_3(conv5_3(temp5_2), bn_attrs));
 
     d_attrs["dropout_prob"] = 0.5f;
-    auto temp6 = bn6(fc1(dropout(pool5, d_attrs)), bn_attrs);
-    return fc3(fc2(dropout(temp6, d_attrs)));
+    auto temp6 = bn6(fc1({dropout(pool5, d_attrs)}), bn_attrs);
+    return fc3({fc2({dropout(temp6, d_attrs)})});
   };
 
   int total_steps = 100000;
   int test_steps = 1000;
   int print_step = 100;
-  float threshold = 0.8f;
+  float threshold = 0.3f;
 
   int iter_num = 1050;
   int skip_batch_num = 50;
   bool model_saved = false;
-  bool do_benchmark = true;
+  bool do_benchmark = false;
 
   auto start = std::chrono::system_clock::now();
   int num_samples = 0;
@@ -332,8 +332,8 @@ TEST(Cifar, TestGPU) {
     auto pool5 = pool2d(inf_bn5_3(inf_conv5_3(temp5_2), bn_attrs));
 
     d_attrs["dropout_prob"] = 0.5f;
-    auto temp6 = inf_bn6(inf_fc1(dropout(pool5, d_attrs)), bn_attrs);
-    return inf_fc3(inf_fc2(dropout(temp6, d_attrs)));
+    auto temp6 = inf_bn6(inf_fc1({dropout(pool5, d_attrs)}), bn_attrs);
+    return inf_fc3({inf_fc2({dropout(temp6, d_attrs)})});
   };
 
   std::vector<float> losses;

src/example/machine_translation/CMakeLists.txt

@@ -0,0 +1,18 @@
+# Copyright (c) 2016 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.
+#
+
+cc_test(test_machine_translation
+        SRCS test_machine_translation.cc
+        DEPS tape tape_function tape_optimizer)

src/example/machine_translation/create_wmt14_recordio.py

@@ -0,0 +1,63 @@
+#   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 paddle
+import paddle.fluid as fluid
+import paddle.dataset.wmt14 as wmt14
+
+
+def create_wmt14_recordio_files():
+    with fluid.program_guard(fluid.Program(), fluid.Program()):
+        reader = paddle.batch(wmt14.train(30000), batch_size=1)
+        feeder = fluid.DataFeeder(
+            feed_list=[
+                fluid.layers.data(
+                    name='src_word_id', shape=[1], dtype='int64', lod_level=1),
+                fluid.layers.data(
+                    name='target_language_word',
+                    shape=[1],
+                    dtype='int64',
+                    lod_level=1), fluid.layers.data(
+                        name='target_language_next_word',
+                        shape=[1],
+                        dtype='int64',
+                        lod_level=1)
+            ],
+            place=fluid.CPUPlace())
+        fluid.recordio_writer.convert_reader_to_recordio_file(
+            '/tmp/wmt14_train.recordio', reader, feeder)
+
+    with fluid.program_guard(fluid.Program(), fluid.Program()):
+        reader = paddle.batch(wmt14.test(30000), batch_size=1)
+        feeder = fluid.DataFeeder(
+            feed_list=[
+                fluid.layers.data(
+                    name='src_word_id', shape=[1], dtype='int64', lod_level=1),
+                fluid.layers.data(
+                    name='target_language_word',
+                    shape=[1],
+                    dtype='int64',
+                    lod_level=1), fluid.layers.data(
+                        name='target_language_next_word',
+                        shape=[1],
+                        dtype='int64',
+                        lod_level=1)
+            ],
+            place=fluid.CPUPlace())
+        fluid.recordio_writer.convert_reader_to_recordio_file(
+            '/tmp/wmt14_test.recordio', reader, feeder)
+
+
+if __name__ == "__main__":
+    create_wmt14_recordio_files()