How can I find the correct input size of the LSTM first layer in keras

Question

I am trying to set the correct input shape for first layer of LSTM in keras but it is tough for me to understand what is the correct input_shape

For print(X_train.shape) I get (9600, 64, 64, 1)

For print(y_train.shape) I get (9600, 15)

#Initializing the classifier Network
classifier = Sequential()

#Adding the input LSTM network layer
classifier.add(LSTM(128, input_shape=(64,1), return_sequences=True))
classifier.add(Dropout(0.2))

If you need more information feel free to ask

Answer 1

This data can't pass into an LSTM layer, which expects 3D data. Maybe try tf.keras.layers.ConvLSTM2D after adding a time steps dimension:

import tensorflow as tf

images = tf.random.uniform((10, 1, 224, 224, 1))

classifier = tf.keras.Sequential([
    tf.keras.layers.ConvLSTM2D(8, 
                               kernel_size=(3, 3),
                               input_shape=(1, 224, 224, 1), 
                               return_sequences=True)
])

classifier(images)

   [[[[-3.53521258e-02, -2.02189311e-02, -2.47801729e-02, ...,
           -2.34759413e-03,  4.60262299e-02,  4.76470888e-02],
          [ 1.04620471e-03, -9.23185516e-03,  1.37878451e-02, ...,
           -4.88127321e-02,  4.20494527e-02,  6.06664363e-03],
          [ 1.26057174e-02,  1.07498122e-02, -1.85700115e-02, ...,
           -1.49483923e-02,  1.21065099e-02,  1.71790868e-02]...,