LSTM model: ValueError: Input 0 of layer "lstm" is incompatible with the layer: expected n-CodePudding

I am building a keras model for a multi-class classification problem. my data set has 7 numerical features and 4 labels. I have structured the model as follows:

def create_keras_model():
  initializer = tf.keras.initializers.GlorotNormal()
  return tf.keras.models.Sequential([
                            #tf.keras.layers.Input(shape=(7,)),
                            LSTM(units=20,kernel_initializer = initializer   
                            input_shape=(7,)), 
                            tf.keras.layers.Dense(4,),
                            tf.keras.layers.Softmax(),
                            ])

and when I compile it, got this error:

ValueError: Input 0 of layer "lstm" is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: (None, 7)

What could be the problem here? and how can I fix it?

CodePudding user response：

You need to change intput_shape and use tf.expand_dims() on X to add one dimension at the end, then you can use your model and start training.

tf.keras.layers.LSTM(units=20, input_shape=(7,1))

x = tf.expand_dims(x, axis=-1)

Full code:

import tensorflow as tf
def create_keras_model():
  initializer = tf.keras.initializers.GlorotNormal()
  return tf.keras.models.Sequential([
                            tf.keras.layers.LSTM(units=20,kernel_initializer = initializer,
                                                 input_shape=(7,1)), 
                            tf.keras.layers.Dense(4,),
                            tf.keras.layers.Softmax(),
                            ])
  
x = tf.random.normal((100,7))
y = tf.random.normal((100,4))

x = tf.expand_dims(x, axis=-1)
model = create_keras_model()
model.compile(optimizer='adam', loss = 'categorical_crossentropy')
model.fit(x,y, epochs=2, batch_size=2)

Output:

Epoch 1/2
50/50 [==============================] - 2s 4ms/step - loss: 0.2027
Epoch 2/2
50/50 [==============================] - 0s 4ms/step - loss: 0.1878