Add save inference model to trainer (#12682)

* add save_inference_model for Trainer

* add comment

* update comment

* fix mac compile

* add unit test

* update API.spec

* revert cpplint-cpp-source

paddle/fluid/API.spec

@@ -43,6 +43,7 @@ paddle.fluid.Executor.run ArgSpec(args=['self', 'program', 'feed', 'fetch_list',
 paddle.fluid.global_scope ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.scope_guard ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
 paddle.fluid.Trainer.__init__ ArgSpec(args=['self', 'train_func', 'optimizer_func', 'param_path', 'place', 'parallel', 'checkpoint_config'], varargs=None, keywords=None, defaults=(None, None, False, None))
+paddle.fluid.Trainer.save_inference_model ArgSpec(args=['self', 'param_path', 'feeded_var_names', 'target_var_indexes'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Trainer.save_params ArgSpec(args=['self', 'param_path'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Trainer.stop ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.Trainer.test ArgSpec(args=['self', 'reader', 'feed_order'], varargs=None, keywords=None, defaults=None)

python/paddle/fluid/tests/book/high-level-api/fit_a_line/test_fit_a_line.py

@@ -47,14 +47,14 @@ def train_program():
     loss = fluid.layers.square_error_cost(input=y_predict, label=y)
     avg_loss = fluid.layers.mean(loss)
 
-    return avg_loss
+    return [avg_loss, y_predict]
 
 
 def optimizer_func():
     return fluid.optimizer.SGD(learning_rate=0.001)
 
 
-def train(use_cuda, train_program, params_dirname):
+def train(use_cuda, train_program, params_dirname, inference_model_dirname):
     place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
 
     trainer = fluid.Trainer(
@@ -74,6 +74,8 @@ def train(use_cuda, train_program, params_dirname):
                 '''
                 if params_dirname is not None:
                     trainer.save_params(params_dirname)
+                    trainer.save_inference_model(inference_model_dirname,
+                                                 ['x'], [1])
                 trainer.stop()
 
     trainer.train(
@@ -99,15 +101,55 @@ def infer(use_cuda, inference_program, params_dirname=None):
     print("infer results: ", results[0])
 
 
+def infer_by_saved_model(use_cuda, save_dirname=None):
+    if save_dirname is None:
+        return
+
+    place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
+    exe = fluid.Executor(place)
+
+    inference_scope = fluid.core.Scope()
+    with fluid.scope_guard(inference_scope):
+        # Use fluid.io.load_inference_model to obtain the inference program desc,
+        # the feed_target_names (the names of variables that will be feeded
+        # data using feed operators), and the fetch_targets (variables that
+        # we want to obtain data from using fetch operators).
+        [inference_program, feed_target_names,
+         fetch_targets] = fluid.io.load_inference_model(save_dirname, exe)
+
+        # The input's dimension should be 2-D and the second dim is 13
+        # The input data should be >= 0
+        batch_size = 10
+
+        test_reader = paddle.batch(
+            paddle.dataset.uci_housing.test(), batch_size=batch_size)
+
+        test_data = next(test_reader())
+        test_feat = numpy.array(
+            [data[0] for data in test_data]).astype("float32")
+        test_label = numpy.array(
+            [data[1] for data in test_data]).astype("float32")
+
+        assert feed_target_names[0] == 'x'
+        results = exe.run(inference_program,
+                          feed={feed_target_names[0]: numpy.array(test_feat)},
+                          fetch_list=fetch_targets)
+        print("infer shape: ", results[0].shape)
+        print("infer results: ", results[0])
+        print("ground truth: ", test_label)
+
+
 def main(use_cuda):
     if use_cuda and not fluid.core.is_compiled_with_cuda():
         return
 
     # Directory for saving the trained model
-    params_dirname = "fit_a_line.inference.model"
+    params_dirname = "fit_a_line.model"
+    inference_model_dirname = "fit_a_line.inference_model"
 
-    train(use_cuda, train_program, params_dirname)
+    train(use_cuda, train_program, params_dirname, inference_model_dirname)
     infer(use_cuda, inference_program, params_dirname)
+    infer_by_saved_model(use_cuda, inference_model_dirname)
 
 
 class TestFitALine(unittest.TestCase):

python/paddle/fluid/trainer.py

@@ -431,6 +431,28 @@ class Trainer(object):
             exe = executor.Executor(self.place)
             io.save_persistables(exe, dirname=param_path)
 
+    def save_inference_model(self, param_path, feeded_var_names,
+                             target_var_indexes):
+        """
+        Save model for cpp inference into :code:`param_path`.
+
+        Args:
+            param_path(str): The path to save parameters.
+            feeded_var_names(list(str)): The name of the vars that you
+                need to feed in before run program.
+            target_var_indexes(list(int)): the index of target var that
+                you need to return in trainer.train_func.
+        Returns:
+            None
+        """
+        with self._prog_and_scope_guard():
+            exe = executor.Executor(self.place)
+            target_vars = [
+                self.train_func_outputs[index] for index in target_var_indexes
+            ]
+            io.save_inference_model(param_path, feeded_var_names, target_vars,
+                                    exe)
+
     @contextlib.contextmanager
     def _prog_and_scope_guard(self):
         with framework.program_guard(