keras之lstm

1、预测sin函数

import numpy as npfrom matplotlib import pyplot as pltfrom keras.models import Sequentialfrom keras.layers import Densefrom keras.layers import LSTMfrom keras import optimizersfrom keras.utils import np_utilsTIMESTEP = 10HIDDEN_UNITS = 100BATCH_SIZE = 10EPOCH = 10model_store = 'model\lstm_sin_model.h5'# 数据可视化def draw(data1, data2):    ax = plt.subplot(111)    ax.plot(list(range(len(data1))), data1, 'b-', list(range(len(data1), len(data1) + len(data2))), data2, 'r-')    plt.show()# 获取数据def get_data():    x = np.linspace(start=0, stop=50, num=200, endpoint=False)    y = np.sin(x)    X = []    Y = []    for i in range(y.shape[0] - TIMESTEP - 1):        X.append(y[i:i + TIMESTEP])        Y.append(y[i + TIMESTEP])    X = np.array(X).reshape(-1, TIMESTEP, 1)    Y = np.array(Y).reshape(-1, 1)    return X, Yx, y = get_data()# 模型def lstm_model():    model = Sequential()    model.add(LSTM(HIDDEN_UNITS, input_shape=(TIMESTEP, 1)))    model.add(Dense(1))    return model# 训练模型def train_model():    model = lstm_model()    # model.compile(loss='mean_squared_error',optimizer='adam')    model.compile(loss='mean_squared_error',                  optimizer=optimizers.Adamax(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08))    model.fit(x, y, batch_size=BATCH_SIZE, nb_epoch=EPOCH, verbose=1)    #模型保存    model.save_weights(model_store)# 预测模型def predict_model():    model = lstm_model()    # 模型加载    model.load_weights(model_store)    result = list()    tmp = x[-1]    print('tmp', tmp)    for i in range(200):        p = model.predict(tmp.reshape(1, TIMESTEP, 1))        result.append(p[0])        tmp[:TIMESTEP - 1, :] = tmp[1:TIMESTEP, :]        tmp[TIMESTEP - 1, :] = p[0]    result = np.array(result)    print('prediction:', result)    draw(y, result)if __name__ == '__main__':    train_model()    predict_model()

2、预测字母，如输入‘abc’输出‘d’

import stringimport numpy as npfrom keras.models import Sequentialfrom keras.layers import Densefrom keras.layers import LSTMfrom keras import optimizersfrom keras.utils import np_utilsTIMESTEP = 3HIDDEN_UNITS = 100BATCH_SIZE = 10EPOCH = 10alphabet = string.ascii_letters# create mapping of characters to integers (0-25) and the reversechar_to_int = dict((c, i) for i, c in enumerate(alphabet))int_to_char = dict((i, c) for i, c in enumerate(alphabet))# prepare the dataset of input to output pairs encoded as integersmodel_store = 'model\lstm_char_model.h5'LEN_CHARS = len(alphabet)alphabet_one_hot = np_utils.to_categorical(range(LEN_CHARS))def get_data():    dataX = []    dataY = []    for i in range(0, len(alphabet) - TIMESTEP, 1):        seq_in = alphabet_one_hot[i:i + TIMESTEP]        seq_out = alphabet_one_hot[i + TIMESTEP]        dataX.append(seq_in)        dataY.append(seq_out)    return np.array(dataX), np.array(dataY)x, y = get_data()# 模型def lstm_model():    model = Sequential()    model.add(LSTM(HIDDEN_UNITS, input_shape=(TIMESTEP, LEN_CHARS)))    model.add(Dense(LEN_CHARS))    return model# 训练模型def train_model():    model = lstm_model()    # model.compile(loss='mean_squared_error',optimizer='adam')    model.compile(loss='mean_squared_error',                  optimizer=optimizers.Adamax(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08))    model.fit(x, y, batch_size=BATCH_SIZE, nb_epoch=EPOCH, verbose=1)    # 模型保存    model.save_weights(model_store)# 预测模型def predict_model():    model = lstm_model()    # 模型加载    model.load_weights(model_store)    for i in range(LEN_CHARS - 3):        str_in = alphabet[i:i + TIMESTEP]        str_to_int = [char_to_int[item] for item in str_in]        str_to_one_hot = np.array([alphabet_one_hot[item] for item in str_to_int])        p = model.predict(str_to_one_hot.reshape(-1, TIMESTEP, LEN_CHARS))        p = ''.join([int_to_char[item] for item in np.argmax(p, axis=1)])        print('input:{}->prediction:{}'.format(str_in, p))if __name__ == '__main__':    train_model()    # predict_model() #输出结果#input:abc->prediction:d#input:bcd->prediction:e#input:cde->prediction:f#input:def->prediction:g#input:efg->prediction:h#input:fgh->prediction:i#input:ghi->prediction:j#input:hij->prediction:k#input:ijk->prediction:l#input:jkl->prediction:m