如何为回归问题定义keras的input_shape？

Question

I have a CSV file containing 200000 rows of 5 features samples (200-time steps in 1000 points). For a regression prediction, I was trying to design an LSTM(keras.Sequential()) because it is a time series problem. When I want to design a model in Keras using "tf.keras.Sequential()" , in the first layer, I have an issue of defining the input_shape. I used input_shape=(train_dataset.shape[0],1,train_dataset.shape[1]) but I got an error. The designed model is below:

def build_model():
    model = tf.keras.Sequential([
        layers.Dense(128,activation=tf.nn.relu,input_shape=(train_dataset.shape[0],1,train_dataset.shape[1])),
        layers.Dense(128,activation=tf.nn.relu),
        layers.Dense(1)
    ])
    opt = tf.keras.optimizers.Adam(learning_rate=0.001)
    model.compile(loss='mean_squared_error', optimizer=opt, metrics= ['mean_squared_error'])
    
    return model

Any suggestion would be appreciated

Answer 1

The input_shape should not include the batch dimension. Use input_shape=train_dataset.shape[1:]

def build_model():
    model = tf.keras.Sequential([
        layers.Dense(128,activation=tf.nn.relu,input_shape=train_dataset.shape[1:]),
        layers.Dense(128,activation=tf.nn.relu),
        layers.Dense(1)
    ])
    opt = tf.keras.optimizers.Adam(learning_rate=0.001)
    model.compile(loss='mean_squared_error', optimizer=opt, metrics= ['mean_squared_error'])
    
    return model