Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into roi_pooling

paddle/operators/math/selected_rows_functor.cu

@@ -227,7 +227,6 @@ template struct SelectedRowsAddToTensor<platform::GPUPlace, float>;
 template struct SelectedRowsAddToTensor<platform::GPUPlace, double>;
 template struct SelectedRowsAddToTensor<platform::GPUPlace, int>;
 template struct SelectedRowsAddToTensor<platform::GPUPlace, int64_t>;
-
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle

python/paddle/v2/fluid/layers.py

@@ -418,6 +418,7 @@ def _create_op_func_(op_type):
 _create_op_func_('mean')
 _create_op_func_('mul')
 _create_op_func_('elementwise_add')
+_create_op_func_('elementwise_div')
 _create_op_func_('dropout')
 _create_op_func_('reshape')
 _create_op_func_('sigmoid')
@@ -457,13 +458,14 @@ def concat(input, axis, main_program=None, startup_program=None):
     return out
 
 
-def sums(input, main_program=None, startup_program=None):
+def sums(input, out=None, main_program=None, startup_program=None):
     """
     This function takes in the input and performs the sum operation on it
     and returns that as the output.
     """
     helper = LayerHelper('sum', **locals())
-    out = helper.create_tmp_variable(dtype=helper.input_dtype())
+    if out is None:
+        out = helper.create_tmp_variable(dtype=helper.input_dtype())
     helper.append_op(type='sum', inputs={'X': input}, outputs={'Out': out})
     return out
 
@@ -606,7 +608,7 @@ def square_error_cost(input, label, **kwargs):
     return square_out
 
 
-def accuracy(input, label, k=1, **kwargs):
+def accuracy(input, label, k=1, correct=None, total=None, **kwargs):
     """
     This function computes the accuracy using the input and label.
     The output is the top_k inputs and their indices.
@@ -620,10 +622,11 @@ def accuracy(input, label, k=1, **kwargs):
         outputs={"Out": [topk_out],
                  "Indices": [topk_indices]},
         attrs={"k": k})
-    acc_out_dtype = kwargs.get("out_dtype", "float32")
     acc_out = helper.create_tmp_variable(dtype="float32")
-    correct = helper.create_tmp_variable(dtype="int64")
+    if correct is None:
-    total = helper.create_tmp_variable(dtype="int64")
+        correct = helper.create_tmp_variable(dtype="int64")
+    if total is None:
+        total = helper.create_tmp_variable(dtype="int64")
     helper.append_op(
         type="accuracy",
         inputs={
@@ -1355,6 +1358,19 @@ def lod_rank_table(x, level=0, main_program=None):
     return table
 
 
+def topk(input, k, main_program=None, startup_program=None):
+    helper = LayerHelper('topk', **locals())
+    topk_out = helper.create_tmp_variable(dtype=input.data_type)
+    topk_indices = helper.create_tmp_variable(dtype='int64')
+    helper.append_op(
+        type='top_k',
+        inputs={'X': [input]},
+        outputs={'Out': [topk_out],
+                 'Indices': [topk_indices]},
+        attrs={'k': k})
+    return topk_out, topk_indices
+
+
 def lod_tensor_to_array(x, table, main_program=None):
     """
     This function creates an operator to convert an LOD_Tensor to
@@ -1388,14 +1404,20 @@ def array_to_lod_tensor(x, table, main_program=None):
     return tmp
 
 
-def fill_constant(shape, dtype, value, main_program=None, startup_program=None):
+def fill_constant(shape,
+                  dtype,
+                  value,
+                  out=None,
+                  main_program=None,
+                  startup_program=None):
     """
     This function creates a tensor , with shape as mentioned in the input and
     specified dtype and fills this up with a constant value that
     comes in the input. It also sets the stop_gradient to be True.
     """
     helper = LayerHelper("fill_constant", **locals())
-    out = helper.create_tmp_variable(dtype=dtype)
+    if out is None:
+        out = helper.create_tmp_variable(dtype=dtype)
     helper.append_op(
         type='fill_constant',
         inputs={},

python/paddle/v2/fluid/tests/book/test_image_classification_train.py

@@ -5,7 +5,6 @@ import paddle.v2.fluid.framework as framework
 import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.nets as nets
 import paddle.v2.fluid.evaluator as evaluator
-from paddle.v2.fluid.io import get_inference_program
 from paddle.v2.fluid.executor import Executor
 from paddle.v2.fluid.initializer import XavierInitializer
 from paddle.v2.fluid.optimizer import AdamOptimizer
@@ -110,18 +109,16 @@ avg_cost = layers.mean(x=cost)
 optimizer = AdamOptimizer(learning_rate=0.001)
 opts = optimizer.minimize(avg_cost)
 
-accuracy, acc_out = evaluator.accuracy(input=predict, label=label)
+accuracy = evaluator.Accuracy(input=predict, label=label)
 
 BATCH_SIZE = 128
 PASS_NUM = 1
 
 train_reader = paddle.batch(
     paddle.reader.shuffle(
-        paddle.dataset.cifar.train10(), buf_size=BATCH_SIZE * 10),
+        paddle.dataset.cifar.train10(), buf_size=128 * 10),
     batch_size=BATCH_SIZE)
 
-test_reader = paddle.batch(paddle.dataset.cifar.test10(), batch_size=BATCH_SIZE)
-
 place = core.CPUPlace()
 exe = Executor(place)
 
@@ -147,46 +144,15 @@ for pass_id in range(PASS_NUM):
         outs = exe.run(framework.default_main_program(),
                        feed={"pixel": tensor_img,
                              "label": tensor_y},
-                       fetch_list=[avg_cost, acc_out])
+                       fetch_list=[avg_cost] + accuracy.metrics)
 
         loss = np.array(outs[0])
         acc = np.array(outs[1])
         pass_acc = accuracy.eval(exe)
-
-        batch_id = batch_id + 1
-
-        test_accuracy, test_acc_out = evaluator.accuracy(
-            input=predict, label=label)
-
-        test_target = [avg_cost, test_acc_out] + test_accuracy.states().values()
-        inference_program = get_inference_program(test_target)
-
-        test_accuracy.reset(exe)
-
-        for data in test_reader():
-            x_data = np.array(map(lambda x: x[0].reshape(data_shape),
-                                  data)).astype("float32")
-            y_data = np.array(map(lambda x: x[1], data)).astype("int64")
-            y_data = np.expand_dims(y_data, axis=1)
-
-            tensor_x = core.LoDTensor()
-            tensor_x.set(x_data, place)
-
-            tensor_y = core.LoDTensor()
-            tensor_y.set(y_data, place)
-
-            outs = exe.run(inference_program,
-                           feed={'pixel': tensor_x,
-                                 'label': tensor_y},
-                           fetch_list=[avg_cost, test_acc_out])
-            out = np.array(outs[0])
-            acc = np.array(outs[1])
-
-        test_pass_acc = test_accuracy.eval(exe)
-
         print("pass_id:" + str(pass_id) + " batch_id:" + str(batch_id) +
               " loss:" + str(loss) + " acc:" + str(acc) + " pass_acc:" + str(
-                  pass_acc) + " test_pass_acc:" + str(test_pass_acc))
+                  pass_acc))
+        batch_id = batch_id + 1
 
         if batch_id > 1:
             # this model is slow, so if we can train two mini batch, we think it works properly.

python/paddle/v2/fluid/tests/book/test_recognize_digits_conv.py

@@ -31,7 +31,7 @@ avg_cost = layers.mean(x=cost)
 optimizer = AdamOptimizer(learning_rate=0.01, beta1=0.9, beta2=0.999)
 opts = optimizer.minimize(avg_cost)
 
-accuracy, acc_out = evaluator.accuracy(input=predict, label=label)
+accuracy = evaluator.Accuracy(input=predict, label=label)
 
 BATCH_SIZE = 50
 PASS_NUM = 3
@@ -61,7 +61,7 @@ for pass_id in range(PASS_NUM):
         outs = exe.run(framework.default_main_program(),
                        feed={"pixel": tensor_img,
                              "label": tensor_y},
-                       fetch_list=[avg_cost, acc_out])
+                       fetch_list=[avg_cost] + accuracy.metrics)
         loss = np.array(outs[0])
         acc = np.array(outs[1])
         pass_acc = accuracy.eval(exe)

python/paddle/v2/fluid/tests/book/test_recognize_digits_mlp.py

@@ -36,7 +36,7 @@ avg_cost = layers.mean(x=cost)
 optimizer = MomentumOptimizer(learning_rate=0.001, momentum=0.9)
 opts = optimizer.minimize(avg_cost)
 
-accuracy, acc_out = evaluator.accuracy(input=predict, label=label)
+accuracy = evaluator.Accuracy(input=predict, label=label)
 
 train_reader = paddle.batch(
     paddle.reader.shuffle(
@@ -67,15 +67,14 @@ for pass_id in range(PASS_NUM):
         outs = exe.run(framework.default_main_program(),
                        feed={'x': tensor_x,
                              'y': tensor_y},
-                       fetch_list=[avg_cost, acc_out])
+                       fetch_list=[avg_cost] + accuracy.metrics)
         out = np.array(outs[0])
         acc = np.array(outs[1])
         pass_acc = accuracy.eval(exe)
 
-        test_accuracy, test_acc_out = evaluator.accuracy(
+        test_accuracy = evaluator.Accuracy(input=predict, label=label)
-            input=predict, label=label)
 
-        test_target = [avg_cost, test_acc_out] + test_accuracy.states().values()
+        test_target = [avg_cost] + test_accuracy.metrics + test_accuracy.states
         inference_program = get_inference_program(test_target)
 
         test_accuracy.reset(exe)
@@ -93,7 +92,7 @@ for pass_id in range(PASS_NUM):
             outs = exe.run(inference_program,
                            feed={'x': tensor_x,
                                  'y': tensor_y},
-                           fetch_list=[avg_cost, test_acc_out])
+                           fetch_list=[avg_cost] + test_accuracy.metrics)
             out = np.array(outs[0])
             acc = np.array(outs[1])
 

python/paddle/v2/fluid/tests/book/test_understand_sentiment_conv.py

@@ -32,9 +32,9 @@ def convolution_net(input_dim, class_dim=2, emb_dim=32, hid_dim=32):
     cost = layers.cross_entropy(input=prediction, label=label)
     avg_cost = layers.mean(x=cost)
     adam_optimizer = AdamOptimizer(learning_rate=0.002)
-    opts = adam_optimizer.minimize(avg_cost)
+    adam_optimizer.minimize(avg_cost)
-    accuracy, acc_out = evaluator.accuracy(input=prediction, label=label)
+    accuracy = evaluator.Accuracy(input=prediction, label=label)
-    return avg_cost, accuracy, acc_out
+    return avg_cost, accuracy, accuracy.metrics[0]
 
 
 def to_lodtensor(data, place):

python/paddle/v2/fluid/tests/book/test_understand_sentiment_dynamic_lstm.py

@@ -41,9 +41,9 @@ def stacked_lstm_net(input_dim,
     cost = layers.cross_entropy(input=prediction, label=label)
     avg_cost = layers.mean(x=cost)
     adam_optimizer = AdamOptimizer(learning_rate=0.002)
-    opts = adam_optimizer.minimize(avg_cost)
+    adam_optimizer.minimize(avg_cost)
-    accuracy, acc_out = evaluator.accuracy(input=prediction, label=label)
+    accuracy = evaluator.Accuracy(input=prediction, label=label)
-    return avg_cost, accuracy, acc_out
+    return avg_cost, accuracy, accuracy.metrics[0]
 
 
 def to_lodtensor(data, place):