任务背景

使用 MLP 做时间序列的二分类问题，通过历史股价判断 未来天数 是涨还是跌。

错误提示

ValueError: Shapes (None, 1) and (None, 2) are incompatible

问题解决

将标签的数值 0，1 转化成 类别的 0，1

from tensorflow.keras.utils import to_categorical
y = to_categorical(dataset['binary_target'].values)

具体程序

import matplotlib.pylab as plt
import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
import tensorflow as tf
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.layers import *
from tensorflow.keras.models import *
# read_data 
data = pd.read_csv('./AAPL.csv')
close_price = data.loc[:, 'Adj Close'].tolist()
close_price_diffs = data.loc[:, 'Adj Close'].pct_change()
# generate the dataset 
target = close_price_diffs.apply(lambda x: 1 if x > 0 else 0)
dataset = pd.DataFrame({'close_price': close_price, 'binary_target':target })
# scale the dataset 
scaler = MinMaxScaler()
X_sc = scaler.fit_transform(dataset['close_price'].values.reshape(-1,1))
# y = dataset['binary_target'].values  ## 会报错，因为不是类别
y = to_categorical(dataset['binary_target'].values)  ## 正确写法！！
X_train, X_test, y_train, y_test = train_test_split(X_sc, y, test_size=0.15, random_state=0)

# define the model 
model = Sequential()
model.add(Dense(64, input_dim=X_train.shape[1]))
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Dense(2))
model.add(Activation('softmax'))

reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.9, patience=5, min_lr=0.000001, verbose=1)
model.compile(optimizer='rmsprop', 
              loss='categorical_crossentropy',
              metrics=['accuracy'])
# fit the  model 
history = model.fit(X_train, y_train, 
          epochs = 50, 
          batch_size = 128, 
          verbose=1, 
          validation_data=(X_test, y_test),
          shuffle=True,
          callbacks=[reduce_lr])