Simpily demo, add paddle.default_main_program() and paddle.default_startup_program

- Removed all main_program and startup_program in the demo.
- Using paddle.default_main_program() hides the implementation detail (e.g., using g_main_program) from the user, we can change the implementation in the future much easier.

python/paddle/v2/__init__.py

@@ -37,6 +37,8 @@ import model
 import paddle.trainer.config_parser as cp
 
 __all__ = [
+    'default_startup_program',
+    'default_main_program',
     'optimizer',
     'layer',
     'activation',

python/paddle/v2/fluid/framework.py

@@ -4,7 +4,7 @@ import collections
 import numpy as np
 import copy
 
-__all__ = ['Block', 'Variable', 'Program', 'Operator']
+__all__ = ['Block', 'Variable', 'Program', 'Operator', 'default_startup_program', 'default_main_program']
 
 
 def unique_name(prefix):
@@ -562,3 +562,9 @@ class Parameter(Variable):
 # program is a global instance.
 g_main_program = Program()
 g_startup_program = Program()
+
+def default_startup_program():
+    return g_startup_program
+
+def default_main_program():
+    return g_main_program

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

@@ -2,45 +2,33 @@ import paddle.v2 as paddle
 import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.io import save_persistables, load_persistables
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
 
-startup_program = Program()
-main_program = Program()
 x = layers.data(
     name='x',
     shape=[13],
-    data_type='float32',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='float32')
 
 y_predict = layers.fc(input=x,
                       size=1,
-                      act=None,
-                      main_program=main_program,
-                      startup_program=startup_program)
+                      act=None)
 
 y = layers.data(
     name='y',
     shape=[1],
-    data_type='float32',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='float32')
 
 cost = layers.square_error_cost(
     input=y_predict,
-    label=y,
-    main_program=main_program,
-    startup_program=startup_program)
-avg_cost = layers.mean(
-    x=cost, main_program=main_program, startup_program=startup_program)
+    label=y)
+avg_cost = layers.mean(x=cost)
 
 sgd_optimizer = optimizer.SGDOptimizer(learning_rate=0.001)
-opts = sgd_optimizer.minimize(avg_cost, startup_program)
+opts = sgd_optimizer.minimize(avg_cost)
 
 BATCH_SIZE = 20
 
@@ -52,12 +40,12 @@ train_reader = paddle.batch(
 place = core.CPUPlace()
 exe = Executor(place)
 
-exe.run(startup_program, feed={}, fetch_list=[])
+exe.run(framework.default_startup_program())
 
 PASS_NUM = 100
 for pass_id in range(PASS_NUM):
-    save_persistables(exe, "./fit_a_line.model/", main_program=main_program)
-    load_persistables(exe, "./fit_a_line.model/", main_program=main_program)
+    save_persistables(exe, "./fit_a_line.model/")
+    load_persistables(exe, "./fit_a_line.model/")
     for data in train_reader():
         x_data = np.array(map(lambda x: x[0], data)).astype("float32")
         y_data = np.array(map(lambda x: x[1], data)).astype("float32")
@@ -69,7 +57,7 @@ for pass_id in range(PASS_NUM):
         tensor_y = core.LoDTensor()
         tensor_y.set(y_data, place)
         # print tensor_y.get_dims()
-        outs = exe.run(main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={'x': tensor_x,
                              'y': tensor_y},
                        fetch_list=[avg_cost])

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

@@ -5,19 +5,17 @@ import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.nets as nets
 import paddle.v2.fluid.optimizer as optimizer
 from paddle.v2.fluid.executor import Executor
-from paddle.v2.fluid.framework import g_startup_program, g_main_program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.initializer import XavierInitializer
 
 
-def resnet_cifar10(input, depth=32, main_program=None, startup_program=None):
+def resnet_cifar10(input, depth=32):
     def conv_bn_layer(input,
                       ch_out,
                       filter_size,
                       stride,
                       padding,
-                      act='relu',
-                      main_program=None,
-                      startup_program=None):
+                      act='relu'):
         tmp = layers.conv2d(
             input=input,
             filter_size=filter_size,
@@ -25,14 +23,10 @@ def resnet_cifar10(input, depth=32, main_program=None, startup_program=None):
             stride=stride,
             padding=padding,
             act=None,
-            bias_attr=False,
-            main_program=main_program,
-            startup_program=startup_program)
+            bias_attr=False)
         return layers.batch_norm(
             input=tmp,
-            act=act,
-            main_program=main_program,
-            startup_program=startup_program)
+            act=act)
 
     def shortcut(input, ch_in, ch_out, stride, program, init_program):
         if ch_in != ch_out:
@@ -44,40 +38,30 @@ def resnet_cifar10(input, depth=32, main_program=None, startup_program=None):
     def basicblock(input,
                    ch_in,
                    ch_out,
-                   stride,
-                   main_program=main_program,
-                   startup_program=startup_program):
+                   stride):
         tmp = conv_bn_layer(
             input,
             ch_out,
             3,
             stride,
-            1,
-            main_program=main_program,
-            startup_program=startup_program)
+            1)
         tmp = conv_bn_layer(
             tmp,
             ch_out,
             3,
             1,
             1,
-            act=None,
-            main_program=main_program,
-            startup_program=startup_program)
-        short = shortcut(input, ch_in, ch_out, stride, main_program,
-                         startup_program)
+            act=None)
+        short = shortcut(input, ch_in, ch_out, stride)
         return layers.elementwise_add(
             x=tmp,
             y=short,
-            act='relu',
-            main_program=main_program,
-            startup_program=startup_program)
+            act='relu')
 
-    def layer_warp(block_func, input, ch_in, ch_out, count, stride, program,
-                   startup_program):
-        tmp = block_func(input, ch_in, ch_out, stride, program, startup_program)
+    def layer_warp(block_func, input, ch_in, ch_out, count, stride):
+        tmp = block_func(input, ch_in, ch_out, stride)
         for i in range(1, count):
-            tmp = block_func(tmp, ch_out, ch_out, 1, program, startup_program)
+            tmp = block_func(tmp, ch_out, ch_out, 1)
         return tmp
 
     assert (depth - 2) % 6 == 0
@@ -87,53 +71,41 @@ def resnet_cifar10(input, depth=32, main_program=None, startup_program=None):
         ch_out=16,
         filter_size=3,
         stride=1,
-        padding=1,
-        main_program=main_program,
-        startup_program=startup_program)
+        padding=1)
     res1 = layer_warp(
         basicblock,
         conv1,
         16,
         16,
         n,
-        1,
-        main_program=main_program,
-        startup_program=startup_program)
+        1)
     res2 = layer_warp(
         basicblock,
         res1,
         16,
         32,
         n,
-        2,
-        main_program=main_program,
-        startup_program=startup_program)
+        2)
     res3 = layer_warp(
         basicblock,
         res2,
         32,
         64,
         n,
-        2,
-        main_program=main_program,
-        startup_program=startup_program)
+        2)
     pool = layers.pool2d(
         input=res3,
         pool_size=8,
         pool_type='avg',
-        pool_stride=1,
-        main_program=main_program,
-        startup_program=startup_program)
+        pool_stride=1)
     return pool
 
 
-def vgg16_bn_drop(input, main_program=None, startup_program=None):
+def vgg16_bn_drop(input):
     def conv_block(input,
                    num_filter,
                    groups,
-                   dropouts,
-                   main_program=None,
-                   startup_program=None):
+                   dropouts):
         return nets.img_conv_group(
             input=input,
             pool_size=2,
@@ -143,51 +115,34 @@ def vgg16_bn_drop(input, main_program=None, startup_program=None):
             conv_act='relu',
             conv_with_batchnorm=True,
             conv_batchnorm_drop_rate=dropouts,
-            pool_type='max',
-            main_program=main_program,
-            startup_program=startup_program)
+            pool_type='max')
 
-    conv1 = conv_block(input, 64, 2, [0.3, 0], main_program, startup_program)
-    conv2 = conv_block(conv1, 128, 2, [0.4, 0], main_program, startup_program)
-    conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0], main_program,
-                       startup_program)
-    conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0], main_program,
-                       startup_program)
-    conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0], main_program,
-                       startup_program)
+    conv1 = conv_block(input, 64, 2, [0.3, 0])
+    conv2 = conv_block(conv1, 128, 2, [0.4, 0])
+    conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
+    conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
+    conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
 
     drop = layers.dropout(
         x=conv5,
-        dropout_prob=0.5,
-        main_program=main_program,
-        startup_program=startup_program)
+        dropout_prob=0.5)
     fc1 = layers.fc(input=drop,
                     size=512,
                     act=None,
-                    param_attr={"initializer": XavierInitializer()},
-                    main_program=main_program,
-                    startup_program=startup_program)
+                    param_attr={"initializer": XavierInitializer()})
     reshape1 = layers.reshape(
         x=fc1,
-        shape=list(fc1.shape + (1, 1)),
-        main_program=main_program,
-        startup_program=startup_program)
+        shape=list(fc1.shape + (1, 1)))
     bn = layers.batch_norm(
         input=reshape1,
-        act='relu',
-        main_program=main_program,
-        startup_program=startup_program)
+        act='relu')
     drop2 = layers.dropout(
         x=bn,
-        dropout_prob=0.5,
-        main_program=main_program,
-        startup_program=startup_program)
+        dropout_prob=0.5)
     fc2 = layers.fc(input=drop2,
                     size=512,
                     act=None,
-                    param_attr={"initializer": XavierInitializer()},
-                    main_program=main_program,
-                    startup_program=startup_program)
+                    param_attr={"initializer": XavierInitializer()})
     return fc2
 
 
@@ -225,7 +180,7 @@ train_reader = paddle.batch(
 place = core.CPUPlace()
 exe = Executor(place)
 
-exe.run(g_startup_program, feed={}, fetch_list=[])
+exe.run(framework.default_startup_program())
 
 for pass_id in range(PASS_NUM):
     batch_id = 0
@@ -243,7 +198,7 @@ for pass_id in range(PASS_NUM):
         tensor_img.set(img_data, place)
         tensor_y.set(y_data, place)
 
-        outs = exe.run(g_main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={"pixel": tensor_img,
                              "label": tensor_y},
                        fetch_list=[avg_cost, accuracy])

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

@@ -3,67 +3,49 @@ import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.nets as nets
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
 
-startup_program = Program()
-main_program = Program()
-
 images = layers.data(
     name='pixel',
     shape=[1, 28, 28],
-    data_type='float32',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='float32')
 label = layers.data(
     name='label',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 conv_pool_1 = nets.simple_img_conv_pool(
     input=images,
     filter_size=5,
     num_filters=20,
     pool_size=2,
     pool_stride=2,
-    act="relu",
-    main_program=main_program,
-    startup_program=startup_program)
+    act="relu")
 conv_pool_2 = nets.simple_img_conv_pool(
     input=conv_pool_1,
     filter_size=5,
     num_filters=50,
     pool_size=2,
     pool_stride=2,
-    act="relu",
-    main_program=main_program,
-    startup_program=startup_program)
+    act="relu")
 
 predict = layers.fc(input=conv_pool_2,
                     size=10,
-                    act="softmax",
-                    main_program=main_program,
-                    startup_program=startup_program)
+                    act="softmax")
 cost = layers.cross_entropy(
     input=predict,
-    label=label,
-    main_program=main_program,
-    startup_program=startup_program)
-avg_cost = layers.mean(x=cost, main_program=main_program)
+    label=label)
+avg_cost = layers.mean(x=cost)
 accuracy = layers.accuracy(
     input=predict,
-    label=label,
-    main_program=main_program,
-    startup_program=startup_program)
+    label=label)
 
 # optimizer = optimizer.MomentumOptimizer(learning_rate=0.1 / 128.0,
 # momentum=0.9)
 optimizer = optimizer.AdamOptimizer(learning_rate=0.01, beta1=0.9, beta2=0.999)
-opts = optimizer.minimize(avg_cost, startup_program)
+opts = optimizer.minimize(avg_cost)
 
 BATCH_SIZE = 50
 PASS_NUM = 3
@@ -75,7 +57,7 @@ train_reader = paddle.batch(
 place = core.CPUPlace()
 exe = Executor(place)
 
-exe.run(startup_program, feed={}, fetch_list=[])
+exe.run(framework.default_startup_program())
 
 for pass_id in range(PASS_NUM):
     count = 0
@@ -90,7 +72,7 @@ for pass_id in range(PASS_NUM):
         tensor_img.set(img_data, place)
         tensor_y.set(y_data, place)
 
-        outs = exe.run(main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={"pixel": tensor_img,
                              "label": tensor_y},
                        fetch_list=[avg_cost, accuracy])

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

@@ -2,8 +2,7 @@ import paddle.v2 as paddle
 import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 from paddle.v2.fluid.regularizer import L2DecayRegularizer
 from paddle.v2.fluid.initializer import UniformInitializer
@@ -11,14 +10,10 @@ from paddle.v2.fluid.initializer import UniformInitializer
 import numpy as np
 
 BATCH_SIZE = 128
-startup_program = Program()
-main_program = Program()
 image = layers.data(
     name='x',
     shape=[784],
-    data_type='float32',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='float32')
 
 param_attr = {
     'name': None,
@@ -30,45 +25,30 @@ param_attr = {
 hidden1 = layers.fc(input=image,
                     size=128,
                     act='relu',
-                    main_program=main_program,
-                    startup_program=startup_program,
                     param_attr=param_attr)
 hidden2 = layers.fc(input=hidden1,
                     size=64,
                     act='relu',
-                    main_program=main_program,
-                    startup_program=startup_program,
                     param_attr=param_attr)
 
 predict = layers.fc(input=hidden2,
                     size=10,
                     act='softmax',
-                    main_program=main_program,
-                    startup_program=startup_program,
                     param_attr=param_attr)
 
 label = layers.data(
     name='y',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 
-cost = layers.cross_entropy(
-    input=predict,
-    label=label,
-    main_program=main_program,
-    startup_program=startup_program)
-avg_cost = layers.mean(
-    x=cost, main_program=main_program, startup_program=startup_program)
+cost = layers.cross_entropy(input=predict, label=label)
+avg_cost = layers.mean(x=cost)
 accuracy = layers.accuracy(
     input=predict,
-    label=label,
-    main_program=main_program,
-    startup_program=startup_program)
+    label=label)
 
 optimizer = optimizer.MomentumOptimizer(learning_rate=0.001, momentum=0.9)
-opts = optimizer.minimize(avg_cost, startup_program)
+opts = optimizer.minimize(avg_cost)
 
 train_reader = paddle.batch(
     paddle.reader.shuffle(
@@ -78,7 +58,7 @@ train_reader = paddle.batch(
 place = core.CPUPlace()
 exe = Executor(place)
 
-exe.run(startup_program, feed={}, fetch_list=[])
+exe.run(framework.default_startup_program())
 
 PASS_NUM = 100
 for pass_id in range(PASS_NUM):
@@ -93,7 +73,7 @@ for pass_id in range(PASS_NUM):
         tensor_y = core.LoDTensor()
         tensor_y.set(y_data, place)
 
-        outs = exe.run(main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={'x': tensor_x,
                              'y': tensor_y},
                        fetch_list=[avg_cost, accuracy])

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

@@ -3,8 +3,7 @@ import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.nets as nets
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program, g_main_program, g_startup_program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
@@ -70,7 +69,7 @@ def main():
     place = core.CPUPlace()
     exe = Executor(place)
 
-    exe.run(g_startup_program)
+    exe.run(framework.default_startup_program())
 
     for pass_id in xrange(PASS_NUM):
         for data in train_data():
@@ -82,7 +81,7 @@ def main():
             tensor_label = core.LoDTensor()
             tensor_label.set(label, place)
 
-            outs = exe.run(g_main_program,
+            outs = exe.run(framework.default_main_program(),
                            feed={"words": tensor_words,
                                  "label": tensor_label},
                            fetch_list=[cost, acc])

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

@@ -3,8 +3,7 @@ import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.nets as nets
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program, g_main_program, g_startup_program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
@@ -81,7 +80,7 @@ def main():
     place = core.CPUPlace()
     exe = Executor(place)
 
-    exe.run(g_startup_program)
+    exe.run(framework.default_startup_program())
 
     for pass_id in xrange(PASS_NUM):
         for data in train_data():
@@ -93,7 +92,7 @@ def main():
             tensor_label = core.LoDTensor()
             tensor_label.set(label, place)
 
-            outs = exe.run(g_main_program,
+            outs = exe.run(framework.default_main_program(),
                            feed={"words": tensor_words,
                                  "label": tensor_label},
                            fetch_list=[cost, acc])

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

@@ -2,8 +2,7 @@ import paddle.v2 as paddle
 import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import g_main_program, g_startup_program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
@@ -88,10 +87,10 @@ def main():
     place = core.CPUPlace()
     tensor_words, tensor_label = prepare_feed_data(data, place)
     exe = Executor(place)
-    exe.run(g_startup_program)
+    exe.run(framework.default_startup_program())
 
     while True:
-        outs = exe.run(g_main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={"words": tensor_words,
                              "label": tensor_label},
                        fetch_list=[cost, acc])

python/paddle/v2/fluid/tests/book/test_word2vec.py

@@ -2,20 +2,17 @@ import paddle.v2 as paddle
 import paddle.v2.fluid.layers as layers
 import paddle.v2.fluid.core as core
 import paddle.v2.fluid.optimizer as optimizer
-
-from paddle.v2.fluid.framework import Program
+import paddle.v2.fluid.framework as framework
 from paddle.v2.fluid.executor import Executor
 
 import numpy as np
 
-startup_program = Program()
-main_program = Program()
-
-embed_size = 32
-hidden_size = 256
+PASS_NUM = 100
+EMBED_SIZE = 32
+HIDDEN_SIZE = 256
 N = 5
-batch_size = 32
-is_sparse = True
+BATCH_SIZE = 32
+IS_SPARSE = True
 
 word_dict = paddle.dataset.imikolov.build_dict()
 dict_size = len(word_dict)
@@ -23,97 +20,67 @@ dict_size = len(word_dict)
 first_word = layers.data(
     name='firstw',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 second_word = layers.data(
     name='secondw',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 third_word = layers.data(
     name='thirdw',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 forth_word = layers.data(
     name='forthw',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 next_word = layers.data(
     name='nextw',
     shape=[1],
-    data_type='int64',
-    main_program=main_program,
-    startup_program=startup_program)
+    data_type='int64')
 
 embed_first = layers.embedding(
     input=first_word,
-    size=[dict_size, embed_size],
+    size=[dict_size, EMBED_SIZE],
     data_type='float32',
-    is_sparse=is_sparse,
-    param_attr={'name': 'shared_w'},
-    main_program=main_program,
-    startup_program=startup_program)
+    is_sparse=IS_SPARSE,
+    param_attr={'name': 'shared_w'})
 embed_second = layers.embedding(
     input=second_word,
-    size=[dict_size, embed_size],
+    size=[dict_size, EMBED_SIZE],
     data_type='float32',
-    is_sparse=is_sparse,
-    param_attr={'name': 'shared_w'},
-    main_program=main_program,
-    startup_program=startup_program)
-
+    is_sparse=IS_SPARSE,
+    param_attr={'name': 'shared_w'})
 embed_third = layers.embedding(
     input=third_word,
-    size=[dict_size, embed_size],
+    size=[dict_size, EMBED_SIZE],
     data_type='float32',
-    is_sparse=is_sparse,
-    param_attr={'name': 'shared_w'},
-    main_program=main_program,
-    startup_program=startup_program)
+    is_sparse=IS_SPARSE,
+    param_attr={'name': 'shared_w'})
 embed_forth = layers.embedding(
     input=forth_word,
-    size=[dict_size, embed_size],
+    size=[dict_size, EMBED_SIZE],
     data_type='float32',
-    is_sparse=is_sparse,
-    param_attr={'name': 'shared_w'},
-    main_program=main_program,
-    startup_program=startup_program)
+    is_sparse=IS_SPARSE,
+    param_attr={'name': 'shared_w'})
 
 concat_embed = layers.concat(
     input=[embed_first, embed_second, embed_third, embed_forth],
-    axis=1,
-    main_program=main_program,
-    startup_program=startup_program)
-
+    axis=1)
 hidden1 = layers.fc(input=concat_embed,
-                    size=hidden_size,
-                    act='sigmoid',
-                    main_program=main_program,
-                    startup_program=startup_program)
+                    size=HIDDEN_SIZE,
+                    act='sigmoid')
 predict_word = layers.fc(input=hidden1,
                          size=dict_size,
-                         act='softmax',
-                         main_program=main_program,
-                         startup_program=startup_program)
+                         act='softmax')
 cost = layers.cross_entropy(
     input=predict_word,
-    label=next_word,
-    main_program=main_program,
-    startup_program=startup_program)
-avg_cost = layers.mean(
-    x=cost, main_program=main_program, startup_program=startup_program)
-
+    label=next_word)
+avg_cost = layers.mean(x=cost)
 sgd_optimizer = optimizer.SGDOptimizer(learning_rate=0.001)
-opts = sgd_optimizer.minimize(avg_cost, startup_program)
+opts = sgd_optimizer.minimize(avg_cost)
 
 train_reader = paddle.batch(
-    paddle.dataset.imikolov.train(word_dict, N), batch_size)
+    paddle.dataset.imikolov.train(word_dict, N), BATCH_SIZE)
 
 place = core.CPUPlace()
 exe = Executor(place)
@@ -122,8 +89,8 @@ exe = Executor(place)
 # below exit line.
 exit(0)
 
-exe.run(startup_program, feed={}, fetch_list=[])
-PASS_NUM = 100
+exe.run(framework.default_startup_program())
+
 for pass_id in range(PASS_NUM):
     for data in train_reader():
         input_data = [[data_idx[idx] for data_idx in data] for idx in xrange(5)]
@@ -150,7 +117,7 @@ for pass_id in range(PASS_NUM):
         next_tensor = core.LoDTensor()
         next_tensor.set(next_data, place)
 
-        outs = exe.run(main_program,
+        outs = exe.run(framework.default_main_program(),
                        feed={
                            'firstw': first_tensor,
                            'secondw': second_tensor,