From 4fe0ed2409b6c04ff53a50ca669e7d31f697ea92 Mon Sep 17 00:00:00 2001 From: Li Fuchen Date: Mon, 14 Oct 2019 11:07:58 +0800 Subject: [PATCH] Revert "Unify 1.6 api in 06.understand_sentiment (#815)" (#820) This reverts commit adcdb7b8574e74bcb0009ee3007395d5144f0b98. --- 06.understand_sentiment/README.cn.md | 24 +++++++--------- 06.understand_sentiment/README.md | 26 ++++++++--------- 06.understand_sentiment/index.cn.html | 24 +++++++--------- 06.understand_sentiment/index.html | 26 ++++++++--------- 06.understand_sentiment/train_conv.py | 28 +++++++++---------- 06.understand_sentiment/train_dyn_rnn.py | 24 ++++++++-------- 06.understand_sentiment/train_stacked_lstm.py | 24 ++++++++-------- 7 files changed, 77 insertions(+), 99 deletions(-) diff --git a/06.understand_sentiment/README.cn.md b/06.understand_sentiment/README.cn.md index 84b44d5..58755fb 100755 --- a/06.understand_sentiment/README.cn.md +++ b/06.understand_sentiment/README.cn.md @@ -151,7 +151,7 @@ BATCH_SIZE = 128 #batch的大小 ```python #文本卷积神经网络 def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) conv_3 = fluid.nets.sequence_conv_pool( input=emb, @@ -183,7 +183,7 @@ def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): #计算词向量 - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) #第一层栈 @@ -218,8 +218,8 @@ def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): ```python def inference_program(word_dict): - data = fluid.data( - name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) net = convolution_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM) @@ -235,7 +235,7 @@ def inference_program(word_dict): ```python def train_program(prediction): - label = fluid.data(name="label", shape=[-1,1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -269,12 +269,12 @@ print("Loading IMDB word dict....") word_dict = paddle.dataset.imdb.word_dict() print ("Reading training data....") -train_reader = fluid.io.batch( +train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") -test_reader = fluid.io.batch( +test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) ``` word_dict是一个字典序列，是词和label的对应关系，运行下一行可以看到具体内容： @@ -401,15 +401,11 @@ reviews = [c.split() for c in reviews_str] UNK = word_dict[''] lod = [] -base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod,re],axis = 0) - base_shape.insert(-1, re.shape[0]) + lod.append([word_dict.get(words, UNK) for words in c]) + +base_shape = [[len(c) for c in lod]] -base_shape = [base_shape] -lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) ``` diff --git a/06.understand_sentiment/README.md b/06.understand_sentiment/README.md index c7db4f6..3d8eae5 100644 --- a/06.understand_sentiment/README.md +++ b/06.understand_sentiment/README.md @@ -140,7 +140,7 @@ Note that `fluid.nets.sequence_conv_pool` contains both convolution and pooling ```python #Textconvolution neural network def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) conv_3 = fluid.nets.sequence_conv_pool( input=emb, @@ -172,7 +172,7 @@ The code of the stack bidirectional LSTM `stacked_lstm_net` is as follows: def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): # Calculate word vectorvector - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) #First stack @@ -191,7 +191,7 @@ def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): inputs = [fc, lstm] #pooling layer - fc_last = fluid.layers.sequence_pool(input=inputs[0], pool_type='max') + pc_last = fluid.layers.sequence_pool(input=inputs[0], pool_type='max') lstm_last = fluid.layers.sequence_pool(input=inputs[1], pool_type='max') #Fully connected layer, softmax prediction @@ -207,8 +207,8 @@ Next we define the prediction program (`inference_program`). We use `convolution ```python def inference_program(word_dict): - data = fluid.data( - name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) net = convolution_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM) @@ -224,7 +224,7 @@ During the testing, the classifier calculates the probability of each output. Th ```python def train_program(prediction): - label = fluid.data(name="label", shape=[-1, 1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -258,12 +258,12 @@ print("Loading IMDB word dict....") word_dict = paddle.dataset.imdb.word_dict() print ("Reading training data....") -train_reader = fluid.io.batch( +train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") -test_reader = fluid.io.batch( +test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) ``` Word_dict is a dictionary sequence, which is the correspondence between words and labels. You can see it specifically by running the next code: @@ -390,15 +390,11 @@ reviews = [c.split() for c in reviews_str] UNK = word_dict[''] lod = [] -base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod,re],axis = 0) - base_shape.insert(-1, re.shape[0]) + lod.append([word_dict.get(words, UNK) for words in c]) + +base_shape = [[len(c) for c in lod]] -base_shape = [base_shape] -lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) ``` diff --git a/06.understand_sentiment/index.cn.html b/06.understand_sentiment/index.cn.html index 3c8645c..d73442f 100644 --- a/06.understand_sentiment/index.cn.html +++ b/06.understand_sentiment/index.cn.html @@ -193,7 +193,7 @@ BATCH_SIZE = 128 #batch的大小 ```python #文本卷积神经网络 def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) conv_3 = fluid.nets.sequence_conv_pool( input=emb, @@ -225,7 +225,7 @@ def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): #计算词向量 - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) #第一层栈 @@ -260,8 +260,8 @@ def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): ```python def inference_program(word_dict): - data = fluid.data( - name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) net = convolution_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM) @@ -277,7 +277,7 @@ def inference_program(word_dict): ```python def train_program(prediction): - label = fluid.data(name="label", shape=[-1,1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -311,12 +311,12 @@ print("Loading IMDB word dict....") word_dict = paddle.dataset.imdb.word_dict() print ("Reading training data....") -train_reader = fluid.io.batch( +train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") -test_reader = fluid.io.batch( +test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) ``` word_dict是一个字典序列，是词和label的对应关系，运行下一行可以看到具体内容： @@ -443,15 +443,11 @@ reviews = [c.split() for c in reviews_str] UNK = word_dict[''] lod = [] -base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod,re],axis = 0) - base_shape.insert(-1, re.shape[0]) + lod.append([word_dict.get(words, UNK) for words in c]) + +base_shape = [[len(c) for c in lod]] -base_shape = [base_shape] -lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) ``` diff --git a/06.understand_sentiment/index.html b/06.understand_sentiment/index.html index eef42a8..e8d3125 100644 --- a/06.understand_sentiment/index.html +++ b/06.understand_sentiment/index.html @@ -182,7 +182,7 @@ Note that `fluid.nets.sequence_conv_pool` contains both convolution and pooling ```python #Textconvolution neural network def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) conv_3 = fluid.nets.sequence_conv_pool( input=emb, @@ -214,7 +214,7 @@ The code of the stack bidirectional LSTM `stacked_lstm_net` is as follows: def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): # Calculate word vectorvector - emb = fluid.embedding( + emb = fluid.layers.embedding( input=data, size=[input_dim, emb_dim], is_sparse=True) #First stack @@ -233,7 +233,7 @@ def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): inputs = [fc, lstm] #pooling layer - fc_last = fluid.layers.sequence_pool(input=inputs[0], pool_type='max') + pc_last = fluid.layers.sequence_pool(input=inputs[0], pool_type='max') lstm_last = fluid.layers.sequence_pool(input=inputs[1], pool_type='max') #Fully connected layer, softmax prediction @@ -249,8 +249,8 @@ Next we define the prediction program (`inference_program`). We use `convolution ```python def inference_program(word_dict): - data = fluid.data( - name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) net = convolution_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM) @@ -266,7 +266,7 @@ During the testing, the classifier calculates the probability of each output. Th ```python def train_program(prediction): - label = fluid.data(name="label", shape=[-1, 1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -300,12 +300,12 @@ print("Loading IMDB word dict....") word_dict = paddle.dataset.imdb.word_dict() print ("Reading training data....") -train_reader = fluid.io.batch( +train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") -test_reader = fluid.io.batch( +test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) ``` Word_dict is a dictionary sequence, which is the correspondence between words and labels. You can see it specifically by running the next code: @@ -432,15 +432,11 @@ reviews = [c.split() for c in reviews_str] UNK = word_dict[''] lod = [] -base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod,re],axis = 0) - base_shape.insert(-1, re.shape[0]) + lod.append([word_dict.get(words, UNK) for words in c]) + +base_shape = [[len(c) for c in lod]] -base_shape = [base_shape] -lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) ``` diff --git a/06.understand_sentiment/train_conv.py b/06.understand_sentiment/train_conv.py index d378ef3..bdc43a2 100644 --- a/06.understand_sentiment/train_conv.py +++ b/06.understand_sentiment/train_conv.py @@ -42,7 +42,8 @@ def parse_args(): def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): - emb = fluid.embedding(input=data, size=[input_dim, emb_dim], is_sparse=True) + emb = fluid.layers.embedding( + input=data, size=[input_dim, emb_dim], is_sparse=True) conv_3 = fluid.nets.sequence_conv_pool( input=emb, num_filters=hid_dim, @@ -61,15 +62,16 @@ def convolution_net(data, input_dim, class_dim, emb_dim, hid_dim): def inference_program(word_dict): - dict_dim = len(word_dict) - data = fluid.data(name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) + dict_dim = len(word_dict) net = convolution_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM) return net def train_program(prediction): - label = fluid.data(name="label", shape=[-1, 1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -88,16 +90,16 @@ def train(use_cuda, params_dirname): print("Reading training data....") if args.enable_ce: - train_reader = fluid.io.batch( + train_reader = paddle.batch( paddle.dataset.imdb.train(word_dict), batch_size=BATCH_SIZE) else: - train_reader = fluid.io.batch( - fluid.io.shuffle( + train_reader = paddle.batch( + paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") - test_reader = fluid.io.batch( + test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) feed_order = ['words', 'label'] @@ -211,15 +213,11 @@ def infer(use_cuda, params_dirname=None): UNK = word_dict[''] lod = [] - base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod, re], axis=0) - base_shape.insert(-1, re.shape[0]) + lod.append([np.int64(word_dict.get(words, UNK)) for words in c]) + + base_shape = [[len(c) for c in lod]] - base_shape = [base_shape] - lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) assert feed_target_names[0] == "words" results = exe.run( diff --git a/06.understand_sentiment/train_dyn_rnn.py b/06.understand_sentiment/train_dyn_rnn.py index 0aec7bb..666bd38 100644 --- a/06.understand_sentiment/train_dyn_rnn.py +++ b/06.understand_sentiment/train_dyn_rnn.py @@ -42,7 +42,8 @@ def parse_args(): def dynamic_rnn_lstm(data, input_dim, class_dim, emb_dim, lstm_size): - emb = fluid.embedding(input=data, size=[input_dim, emb_dim], is_sparse=True) + emb = fluid.layers.embedding( + input=data, size=[input_dim, emb_dim], is_sparse=True) sentence = fluid.layers.fc(input=emb, size=lstm_size * 4, act='tanh') lstm, _ = fluid.layers.dynamic_lstm(sentence, size=lstm_size * 4) @@ -53,7 +54,8 @@ def dynamic_rnn_lstm(data, input_dim, class_dim, emb_dim, lstm_size): def inference_program(word_dict): - data = fluid.data(name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) pred = dynamic_rnn_lstm(data, dict_dim, CLASS_DIM, EMB_DIM, LSTM_SIZE) @@ -61,7 +63,7 @@ def inference_program(word_dict): def train_program(prediction): - label = fluid.data(name="label", shape=[-1, 1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -79,16 +81,16 @@ def train(use_cuda, params_dirname): print("Reading training data....") if args.enable_ce: - train_reader = fluid.io.batch( + train_reader = paddle.batch( paddle.dataset.imdb.train(word_dict), batch_size=BATCH_SIZE) else: - train_reader = fluid.io.batch( + train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") - test_reader = fluid.io.batch( + test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) feed_order = ['words', 'label'] @@ -200,15 +202,11 @@ def infer(use_cuda, params_dirname=None): UNK = word_dict[''] lod = [] - base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod, re], axis=0) - base_shape.insert(-1, re.shape[0]) + lod.append([np.int64(word_dict.get(words, UNK)) for words in c]) + + base_shape = [[len(c) for c in lod]] - base_shape = [base_shape] - lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) assert feed_target_names[0] == "words" results = exe.run( diff --git a/06.understand_sentiment/train_stacked_lstm.py b/06.understand_sentiment/train_stacked_lstm.py index 1f8eb7c..acf38b2 100644 --- a/06.understand_sentiment/train_stacked_lstm.py +++ b/06.understand_sentiment/train_stacked_lstm.py @@ -46,7 +46,8 @@ def parse_args(): def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): assert stacked_num % 2 == 1 - emb = fluid.embedding(input=data, size=[input_dim, emb_dim], is_sparse=True) + emb = fluid.layers.embedding( + input=data, size=[input_dim, emb_dim], is_sparse=True) fc1 = fluid.layers.fc(input=emb, size=hid_dim) lstm1, cell1 = fluid.layers.dynamic_lstm(input=fc1, size=hid_dim) @@ -68,7 +69,8 @@ def stacked_lstm_net(data, input_dim, class_dim, emb_dim, hid_dim, stacked_num): def inference_program(word_dict): - data = fluid.data(name="words", shape=[-1], dtype="int64", lod_level=1) + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) dict_dim = len(word_dict) net = stacked_lstm_net(data, dict_dim, CLASS_DIM, EMB_DIM, HID_DIM, @@ -78,7 +80,7 @@ def inference_program(word_dict): def train_program(prediction): # prediction = inference_program(word_dict) - label = fluid.data(name="label", shape=[-1, 1], dtype="int64") + label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(cost) accuracy = fluid.layers.accuracy(input=prediction, label=label) @@ -98,16 +100,16 @@ def train(use_cuda, params_dirname): print("Reading training data....") if args.enable_ce: - train_reader = fluid.io.batch( + train_reader = paddle.batch( paddle.dataset.imdb.train(word_dict), batch_size=BATCH_SIZE) else: - train_reader = fluid.io.batch( + train_reader = paddle.batch( paddle.reader.shuffle( paddle.dataset.imdb.train(word_dict), buf_size=25000), batch_size=BATCH_SIZE) print("Reading testing data....") - test_reader = fluid.io.batch( + test_reader = paddle.batch( paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE) feed_order = ['words', 'label'] @@ -221,15 +223,11 @@ def infer(use_cuda, params_dirname=None): UNK = word_dict[''] lod = [] - base_shape = [] - for c in reviews: - re = np.array([np.int64(word_dict.get(words, UNK)) for words in c]) - lod = np.concatenate([lod, re], axis=0) - base_shape.insert(-1, re.shape[0]) + lod.append([np.int64(word_dict.get(words, UNK)) for words in c]) + + base_shape = [[len(c) for c in lod]] - base_shape = [base_shape] - lod = np.array(lod).astype('int64') tensor_words = fluid.create_lod_tensor(lod, base_shape, place) assert feed_target_names[0] == "words" results = exe.run( -- GitLab