CODEWITHSUNDEEP
tensorflowkerasdeep-learninglstmrecurrent-neural-network
Rachid
Rachid

Reputation: 39

ValueError: Input 0 is incompatible with layer lstm_60: expected ndim=3, found ndim=2

I want to build a deep RNN where my x_train shape is (318,39) and my y_train has shape (318,). When I execute the code below:

model.add(LSTM(20,input_shape=(X_train.shape[1:]), activation='relu', return_sequences=True))
model.add(LSTM(20, activation='relu'))
model.add(Dense(10, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer='adam',loss='binary_crossentropy',metrics=['accuracy'])
history = model.fit(X_train,y_train,batch_size=20,epochs=250)

I'm getting following error:

ValueError: Input 0 is incompatible with layer lstm_60: expected ndim=3, found ndim=2

Upvotes: 1

Views: 451

Answers (3)

Zabir Al Nazi Nabil
Zabir Al Nazi Nabil

Reputation: 11198

Just reshape the X_train

X_train.reshape(X_train.shape[0],X_train.shape[1],1)

before fit function.

LSTM needs a 3D array (batch size, timesteps, features). In your case, as you have 1 feature, you need to add one extra dimension.

Upvotes: 1

Bashir Kazimi
Bashir Kazimi

Reputation: 1377

Since you are using LSTM, I am assuming your input data is sequential, i.e., you have 318 examples where each example has 39 time steps? If that is the case, you should first reshape your input data properly such as:

import numpy as np
X_train = np.expand_dims(X_train, -1)

This will reshape your X_train to have a shape of (318, 39, 1) and then it will work (Only if my initial assumption is correct)

Upvotes: 2

Vlad
Vlad

Reputation: 8585

The expected input shape of the LSTM layer is [batch, timesteps, feature]. You're passing [batch, timesteps]. What you want to do is to pass [batch, timesteps, 1] (expand dimension on the right). You could do it like this:

X_train = X_train[..., None]

Upvotes: 1

Related Questions