keras 基础入门整理

import keras.preprocessing.text as T
from keras.preprocessing.text import Tokenizer
import keras
print(keras.__version__)
# text1='some thing to eat'
# text2='some thing to drink'
# text是经过分词处理的语句,分隔符要与Tokenizer中split选择相同
text1='我 爱 北京'
text2='我 爱 北京 天安门'
texts=[text1,text2]

print (T.text_to_word_sequence(text1))  #['some', 'thing', 'to', 'eat']
#基于hash函数(桶大小为n,n: int. Size of vocabulary)，将一行文本转换向量表示
print (T.one_hot(text1,n=5))  #[7, 9, 3, 4]
print (T.one_hot(text2,n=10))  #[7, 9, 3, 1]
#num_words为词的总长度，这里暂定为10
tokenizer = Tokenizer(num_words=10,split=" ")#num_words:None或整数,处理的最大单词数量。少于此数的单词丢掉

# tokenizer.fit_on_text(texts)
tokenizer.fit_on_texts(texts)

# print (tokenizer.word_count )#[('some', 2), ('thing', 2), ('to', 2), ('eat', 1), ('drink', 1)]
print(tokenizer.word_counts)
print (tokenizer.word_index )#{'some': 1, 'thing': 2,'to': 3 ','eat': 4, drink': 5}
print (tokenizer.word_docs) #{'some': 2, 'thing': 2, 'to': 2, 'drink': 1,  'eat': 1}
print (tokenizer.index_docs )#{1: 2, 2: 2, 3: 2, 4: 1, 5: 1}
print("#"*5+"转化词向量"+"#"*5)
# print (tokenizer.text_to_sequences(texts) )#[[1, 2, 3, 4], [1, 2, 3, 5]]

# num_words=多少会影响下面的结果，行数=num_words

print (tokenizer.texts_to_sequences(texts) )#[[1, 2, 3, 4], [1, 2, 3, 5]]
print (tokenizer.texts_to_matrix(texts) )#

2.2.2
['我', '爱', '北京']
[4, 1, 2]
[9, 6, 7, 4]
OrderedDict([('我', 2), ('爱', 2), ('北京', 2), ('天安门', 1)])
{'我': 1, '爱': 2, '北京': 3, '天安门': 4}
defaultdict(<class 'int'>, {'北京': 2, '我': 2, '爱': 2, '天安门': 1})
defaultdict(<class 'int'>, {3: 2, 1: 2, 2: 2, 4: 1})
#####转化词向量#####
[[1, 2, 3], [1, 2, 3, 4]]
[[0. 1. 1. 1. 0. 0. 0. 0. 0. 0.]
 [0. 1. 1. 1. 1. 0. 0. 0. 0. 0.]]

save_model(model,path)
load_model(path)

# coding=utf-8
import keras
from keras.models import Sequential
from keras.layers import Dense

from tensorflow.examples.tutorials.mnist import input_data

from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets
# from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets
# mnist =  read_data_sets(one_hot=True)
mnist = read_data_sets("MNIST_data/", one_hot=True)
img_train,label_train = mnist.train.next_batch(60000)
img_test,label_test   = mnist.test.next_batch(10000)

#构建模型
model = Sequential()
model.add(Dense(512,activation='relu',input_dim=784))
model.add(Dense(512,activation='relu'))
model.add(Dense(10,activation='softmax'))
model.compile(optimizer='Adagrad',loss='categorical_crossentropy',metrics=['accuracy'])

#训练及评估模型
hist = model.fit(img_train,label_train)
loss,accu = model.evaluate(img_test,label_test)

print("loss is %f accu is %f"%(loss,accu))

 32/10000 [..............................] - ETA: 12s
  576/10000 [>.............................] - ETA: 1s 
 1152/10000 [==>...........................] - ETA: 1s
 1728/10000 [====>.........................] - ETA: 0s
 2336/10000 [======>.......................] - ETA: 0s
 3008/10000 [========>.....................] - ETA: 0s
 3680/10000 [==========>...................] - ETA: 0s
 4352/10000 [============>.................] - ETA: 0s
 5056/10000 [==============>...............] - ETA: 0s
 5760/10000 [================>.............] - ETA: 0s
 6432/10000 [==================>...........] - ETA: 0s
 7072/10000 [====================>.........] - ETA: 0s
 7776/10000 [======================>.......] - ETA: 0s
 8288/10000 [=======================>......] - ETA: 0s
 8992/10000 [=========================>....] - ETA: 0s
 9664/10000 [===========================>..] - ETA: 0s
10000/10000 [==============================] - 1s 84us/step
loss is 0.079320 accu is 0.975500

import keras
from keras.layers import Dense,Input
from keras.models import Model
from tensorflow.examples.tutorials.mnist import input_data

mnist =  input_data.read_data_sets("MNIST_data/",one_hot=True)

img_train,label_train = mnist.train.next_batch(60000)
img_test,label_test   = mnist.test.next_batch(10000)

input = Input(shape=(784,))
layer1 = Dense(512,activation='relu')(input)
layer2 = Dense(512,activation='relu')(layer1)
pred = Dense(10,activation='softmax')(layer2)

model = Model(inputs=input,outputs=pred)
model.compile(optimizer='rmsprop',loss='categorical_crossentropy',metrics=['accuracy'])

hist = model.fit(img_train,label_train,epochs=1)
loss,accu = model.evaluate(img_test,label_test)
print("loss is %f accu is %f"%(loss,accu))

 32/10000 [..............................] - ETA: 9s
  672/10000 [=>............................] - ETA: 1s
 1248/10000 [==>...........................] - ETA: 0s
 1696/10000 [====>.........................] - ETA: 0s
 2336/10000 [======>.......................] - ETA: 0s
 2976/10000 [=======>......................] - ETA: 0s
 3584/10000 [=========>....................] - ETA: 0s
 4192/10000 [===========>..................] - ETA: 0s
 4800/10000 [=============>................] - ETA: 0s
 5440/10000 [===============>..............] - ETA: 0s
 5984/10000 [================>.............] - ETA: 0s
 6592/10000 [==================>...........] - ETA: 0s
 7200/10000 [====================>.........] - ETA: 0s
 7808/10000 [======================>.......] - ETA: 0s
 8448/10000 [========================>.....] - ETA: 0s
 9024/10000 [==========================>...] - ETA: 0s
 9632/10000 [===========================>..] - ETA: 0s
10000/10000 [==============================] - 1s 89us/step
loss is 0.146895 accu is 0.960400

from keras.callbacks import ModelCheckpoint

model = Sequential()
model.add(Dense(10, input_dim=784, kernel_initializer='uniform'))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')

checkpointer = ModelCheckpoint(filepath="/tmp/weights.h5", save_best_only=True)
tensbrd = TensorBoard(logdir='path/of/log')
model.fit(X_train, Y_train, batch_size=128, callbacks=[checkpointer,tensbrd])

from keras.utils import plot_model
model = keras.applications.InceptionV3()
plot_model(model, to_file='model.png')

import keras
from keras.datasets import cifar10
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Conv2D, MaxPooling2D

num_classes = 10
model_name = 'cifar10.h5'

# The data, shuffled and split between train and test sets:
(x_train, y_train), (x_test, y_test) = cifar10.load_data()

x_train = x_train.astype('float32')/255
x_test = x_test.astype('float32')/255

# Convert class vectors to binary class matrices.
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()

model.add(Conv2D(32, (3, 3), padding='same', input_shape=x_train.shape[1:]))
model.add(Activation('relu'))

model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Conv2D(64, (3, 3), padding='same'))
model.add(Activation('relu'))

model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))

model.add(Dense(num_classes))
model.add(Activation('softmax'))

model.summary()

# initiate RMSprop optimizer
opt = keras.optimizers.rmsprop(lr=0.001, decay=1e-6)

# train the model using RMSprop
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])

hist = model.fit(x_train, y_train, epochs=1, shuffle=True)
model.save(model_name)

# evaluate
loss, accuracy = model.evaluate(x_test, y_test)
print (loss, accuracy)

9664/10000 [===========================>..] - ETA: 0s
 9728/10000 [============================>.] - ETA: 0s
 9792/10000 [============================>.] - ETA: 0s
 9856/10000 [============================>.] - ETA: 0s
 9920/10000 [============================>.] - ETA: 0s
 9984/10000 [============================>.] - ETA: 0s
10000/10000 [==============================] - 9s 853us/step
1.252797156715393 0.5627

model = Sequential()
model.add(Bidirectional(LSTM(10, return_sequences=True), input_shape=(5, 10)))
model.add(Bidirectional(LSTM(10)))
model.add(Dense(5))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')

5  经典作品 道听途说 以为 懂 实际 读 觉得 独特 意想不到 新颖                  
5  东京 看过 热海 看过 回家                  
5  这部 电影 里 看出 东西 太多 比方说 注意 尾道 家中 鸡冠花 会 明白 黑白 影像 彩色 影像 不能 取代 魅力 母亲 热海 防波堤 上 说 东京 游览 热海 回家 真是 人生 非常 隽永 总结                  
5  刚 几幕 觉得 极 做作 哪有 众人 说 这般 好 再 耐心 看 下去 方 发觉 表面 客套 微笑 下 内心深处 悲凉 其实 很 幸福 其实 幸福 等到 老时 会 老伴 相视而笑
5  生活 总是 人 失望                  

texts = []
labels = []
for line in lines:
    fields = line.split()
    rate = int(fields[0])
    if rate==0 or rate==3:
        continue
    elif rate < 3:
        rate = 0
    else:
        rate = 1
    cont = fields[1:]
    cont = " ".join(cont)
    texts.append(cont)
    labels.append(rate)
tokenizer.fit_on_texts(texts)
tokenizer.texts_to_sequence(texts)

model = Sequential()
model.add(Embedding(vocab_size, 256, input_length=sentence_max_len))
model.add(Bidirectional(LSTM(128,implementation=2)))
model.add(Dropout(0.5))
model.add(Dense(2, activation='relu'))
model.compile('RMSprop', 'categorical_crossentropy', metrics=['accuracy'])

from keras.models import Sequential
from keras.preprocessing.text import Tokenizer
import keras.preprocessing.sequence as S
from keras.utils import to_categorical
from keras.layers import Embedding, Bidirectional, LSTM, Dropout, Dense

import jieba
import json
import numpy as np

vocab_size = 350000
sentence_max_len = 100
model_path = 'keras.h5'

class SentimentLSTM:
    def __init__(self):
        self.tokenizer = Tokenizer(num_words=vocab_size)
        self.stop_words = []
        self.model = None

    def load_stop_word(self,path='dict/stop_word.txt'):
        with open(path, 'r') as f:
            for line in f:
                content = line.strip()
                self.stop_words.append(content.decode('utf-8'))

    def jieba_cut(self,line):
        lcut = jieba.lcut(line)
        cut = [x for x in lcut if x not in self.stop_words]
        cut = " ".join(cut)
        return cut

    def load_cuted_corpus(self, dir, input):
        f = open(dir + '/' + input , 'r')
        lines = f.readlines()
        texts = []
        labels = []
        for line in lines:
            fields = line.split()
            rate = int(fields[0])
            if rate==0 or rate==3:
                continue
            elif rate < 3:
                rate = 0
            else:
                rate = 1
            cont = fields[1:]
            cont = " ".join(cont)
            texts.append(cont)
            labels.append(rate)

        self.tokenizer.fit_on_texts(texts)
        f.close()
        return texts,labels

    def load_data(self):
        x,y = self.load_cuted_corpus('corpus', 'review.csv')
        x = self.tokenizer.texts_to_sequences(x)
        x = S.pad_sequences(x,maxlen=sentence_max_len)
        y = to_categorical(y,num_classes=2)
        return ((x[0:500000],y[0:500000]), (x[500000:], y[500000:]))

    def train(self,epochs=50):
        print ('building model ...')
        self.model = SentimentLSTM.build_model()

        print ('loading data ...')
        (text_train, rate_train), (text_test, rate_text) = self.load_data()

        print ('training model ...')
        self.model.fit(text_train, rate_train,batch_size=1000,epochs=epochs)
        self.model.save('model/keras.model')
        score = self.model.evaluate(text_test,rate_text)
        print (score)

    def load_trained_model(self,path):
        model = SentimentLSTM.build_model()
        model.load_weights(path)
        return model

    def predict_text(self,text):
        if self.model == None:
            self.model = self.load_trained_model(model_path)
            self.load_stop_word()
            self.load_cuted_corpus('corpus', 'review.csv')

        vect = self.jieba_cut(text)
        vect = vect.encode('utf-8')
        vect = self.tokenizer.texts_to_sequences([vect,])
        print (vect)
        return self.model.predict_classes(S.pad_sequences(np.array(vect),100))

    @staticmethod
    def build_model():
        model = Sequential()
        model.add(Embedding(vocab_size, 256, input_length=sentence_max_len))
        model.add(Bidirectional(LSTM(128,implementation=2)))
        model.add(Dropout(0.5))
        model.add(Dense(2, activation='relu'))
        model.compile('RMSprop', 'categorical_crossentropy', metrics=['accuracy'])
        return model

def main():
    lstm = SentimentLSTM()
    lstm.train(10)
    while True:
        input = input('Please input text:')
        if input == 'quit':
            break
        print (lstm.predict_text(input))

if __name__=="__main__":
    main()