ValueError：检查输入时出错：预期 lstm_6_input 具有形状 (87482, 1) 但得到的数组具有形状 (87482, 3)

Question

X_train data dimension is 87482, 3.

But I got an error on running with the below code. The error is:

ValueError: Error when checking input: expected lstm_6_input to have shape (87482, 1) but got array with shape (87482, 3)

My code is:

model = Sequential()
#model.add(Embedding(top_words, embedding_vecor_length, input_length=max_review_length))
model.add(LSTM(units=3, input_shape=(X_train_rnn.shape[1],1),return_sequences=True))
model.add(LSTM(3, return_sequences=True))  # returns a sequence of vectors of dimension 32
model.add(LSTM(3))

model.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
print(model.summary())
model.fit(X_train_rnn, y_train, epochs=2, batch_size=32)
# Final evaluation of the model
scores = model.evaluate(X_test_rnn, y_test, verbose=0)

Answer 1

input_shape=(X_train.shape[1], 1) should fix the error, but maybe not the entire problem: how long is each sequence ?

• LSTM full input shape ( batch_shape ) is (batch_size, timesteps, channels) - or, equivalently, (samples, timesteps, features)
• batch_size=32 in your .fit() , but if X_train dimension is (87472, 3) , do you have 87472 samples (sequences) each of length 3 ( timesteps=3 )? If so, you'll need input_shape=(3, 1) (univariate data), which is what you get with (X_train.shape[1], 1)
• If instead your timesteps=87462 , you'll need input_shape=(87462, 3) - but this is a pretty bad idea, as LSTMs struggle for timesteps > 0

I don't know what shapes X_train_rnn has, so I wrote the answer in terms of X_train . Feel free to clarify.