dalian-university/秀水分区/model/model_keras_fenqu.py

# -*- coding: UTF-8 -*-
from keras.layers import Input, Dense, LSTM, concatenate, Conv1D, Conv2D, MaxPooling1D, Reshape, Flatten, Lambda
from keras.models import Model, load_model
from keras.callbacks import ModelCheckpoint, EarlyStopping
from keras import optimizers
from keras.callbacks import TensorBoard
import matplotlib.pyplot as plt
import numpy as np
from keras.callbacks import TensorBoard, EarlyStopping


def get_keras_model(opt):
    lstm_input = Input(shape=(opt.Model['time_step'], opt.input_size_lstm))
    lstm = lstm_input
    for i in range(opt.Model['lstm_layers']):
        rs = True
        if i == opt.Model['lstm_layers']-1:
            rs = False
        lstm = LSTM(units=opt.Model['hidden_size'], dropout=opt.Model['dropout_rate'], return_sequences=rs)(lstm)
    output = Dense(16, name='dense_1')(lstm)
    # output = Flatten(data_format='channels_last')(output)

    lstm1 = lstm_input
    for i in range(opt.Model['lstm_layers']):
        rs = True
        if i == opt.Model['lstm_layers']-1:
            rs = False
        lstm1 = LSTM(units=opt.Model['hidden_size'], dropout=opt.Model['dropout_rate'], return_sequences=rs)(lstm1)
    output1 = Dense(16, name='dense_2')(lstm1)
    # output1 = Flatten(data_format='channels_last')(output1)

    outputs = Lambda(sum)([output, output1])
    # outputs = Dense(16, name='dense_3')(outputs)
    model = Model(lstm_input, [output, output1])
    # model = Model(lstm_input, outputs)
    # model.compile(loss={'dense_1': 'mse', 'dense_2': 'mse', 'dense_3': 'mse'},
    #               loss_weights={'dense_1': 500, 'dense_2': 500, 'dense_3': 0.04},
    #               metrics={'dense_1': ['accuracy', 'mse'], 'dense_2': ['accuracy', 'mse'], 'dense_3': ['accuracy', 'mse']},
    #               optimizer='adam')     # metrics=["mae"]
    model.compile(loss={'dense_1': 'mse', 'dense_2': 'mse'},
                  metrics={'dense_1': ['accuracy', 'mse'], 'dense_2': ['accuracy', 'mse'],},
                  optimizer='adam')  # metrics=["mae"]
    return model


def train_init(use_cuda=False):
    import tensorflow as tf
    from keras.backend.tensorflow_backend import set_session
    if use_cuda:
        # gpu init
        sess_config = tf.ConfigProto(log_device_placement=True, allow_soft_placement=True)
        sess_config.gpu_options.per_process_gpu_memory_fraction = 0.7  # 最多使用70%GPU内存
        sess_config.gpu_options.allow_growth=True   # 初始化时不全部占满GPU显存, 按需分配
        sess = tf.Session(config=sess_config)
        set_session(sess)
    else:
        session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
        tf.set_random_seed(1234)
        sess = tf.Session(graph=tf.get_default_graph(), config=session_conf)
        set_session(sess)


def train(opt, train_and_valid_data):
    train_init(opt.use_cuda)
    train_X, train_Y, valid_X, valid_Y = train_and_valid_data
    print("----------", np.array(train_X[0]).shape)
    print("++++++++++", np.array(train_X[1]).shape)
    model = get_keras_model(opt)
    model.summary()
    weight_lstm_1, bias_lstm_1 = model.get_layer('dense_1').get_weights()
    print("weight_lstm_1 = ", weight_lstm_1)
    print("bias_lstm_1 = ", bias_lstm_1)
    if opt.add_train:
        model.load_weights(opt.model_save_path + 'model_kerass.h5')

    check_point = ModelCheckpoint(filepath=opt.model_save_path + opt.save_name, monitor='val_loss',
                                    save_best_only=True, mode='auto')
    early_stop = EarlyStopping(monitor='val_loss', patience=opt.Model['patience'], mode='auto')

    history = model.fit(train_X, train_Y, batch_size=opt.Model['batch_size'], epochs=opt.Model['epoch'], verbose=2,
              validation_data=(valid_X, valid_Y), callbacks=[check_point, early_stop])
    loss = history.history['loss']
    epochs = range(1, len(loss) + 1)
    plt.title('Loss')
    # plt.plot(epochs, acc, 'red', label='Training acc')
    plt.plot(epochs, loss, 'blue', label='Validation loss')
    plt.legend()
    # plt.show()

def predict(config, test_X):
    model = get_keras_model(config)
    model.load_weights(config.model_save_path + 'model_' + config.save_frame + '.h5')
    result = model.predict(test_X, batch_size=1)
    # result = result.reshape((-1, config.output_size))
    return result