利用tensorflow搭建神经网络

• 获取数据
• 搭建循环神经网络
• 循环神经网络的预测和评估

获取数据

def get_data(code='000001.SZ',start='20180101',end='20191201',window_len=12):
    ts_pro = ts.pro_api('xx')
    df = ts_pro.daily(ts_code=code, start_date=start, end_date=end)
    df = df.reindex(index=df.index[::-1])
    # 转换为二维数组
    arr = df['close'].values
    #归一化处理
    arr=np.reshape(arr,newshape=(-1,1))
    mm=MinMaxScaler()
    arr=mm.fit_transform(arr)
    arr=np.reshape(arr,newshape=-1)
    # 列表[]
    X = []
    Label = []
    for i in range(len(arr) - window_len - 1):
        X.append(arr[i:i + window_len])
        Label.append(arr[i + window_len])
    X=np.array(X)
    Label=np.array(Label)
    train_len = int(len(X) / 3) * 2

    # 划分测试和训练数据集
    X_train = X[:train_len]
    Label_train = Label[:train_len]
    X_test = X[train_len:]
    X_Label = X[train_len:]
    return X_train, Label_train, X_test, X_Label

搭建循环神经网络

indows_len=12
    hidden_units=64
    model_layers=[
        tf.keras.layers.Reshape((windows_len,1),input_shape=(windows_len, )),
        tf.keras.layers.SimpleRNN(units=hidden_units,return_sequences=False),
        tf.keras.layers.Dense(units=hidden_units, activation='sigmoid'),
        tf.keras.layers.Dense(1, activation='sigmoid'),
    ]
    model=tf.keras.models.Sequential(model_layers)
    # print(model.summary())

循环神经网络的预测和评估

def plot_loss(loss):
    fig=plt.figure()
    ax=fig.add_subplot(111)
    ax.plot(range(len(loss)),loss)
    plt.grid(True)
    plt.show()

def plot_result(prediction,true):
    fig=plt.figure()
    ax=fig.add_subplot(111)
    ax.plot(range(len(prediction)),prediction)
    ax.plot(range(len(true)),true)
    plt.legend(['prediction','true'])
    plt.show()

if __name__ == '__main__':
    X_train, Label_train, X_test, X_Label = get_data()
    model.compile(optimizer='adam',loss='mean_squared_error')
    history=model.fit(X_train, Label_train,epochs=100)
    plot_loss(history.history['loss'])
    prediction=model.predict(X_test)
    plot_result(prediction,Lable_test)