CODEWITHSUNDEEP
pythontensorflowkeras
Shamoon
Shamoon

Reputation: 43491

ValueError: Argument must be a dense tensor: ((10, 4945), 1024) - got shape [2], but wanted [2, 2]

My keras model is:

        model = Sequential()
        model.add(Embedding(input_dim=(self.BATCH_SIZE,
                                       len(self._tokens)), output_dim=1024))

        model.add(LSTM(128))

        model.add(Dropout(rate=0.5))
        model.add(Dense(len(self._tokens)))
        model.add(Activation('softmax'))

and I get an error:

ValueError: Argument must be a dense tensor: ((10, 4945), 1024) - got shape [2], but wanted [2, 2].

I'm not sure what I'm doing incorrectly. Any help would be appreciated.

Upvotes: 1

Views: 699

Answers (1)

Anubhav Singh
Anubhav Singh

Reputation: 8699

As per the keras official Documentation, input_dim parameter of embedding layer should be the size of the vocabulary, i.e. maximum integer index + 1 (int>0).

So, you code should be:

model.add(Embedding(input_dim=len(self._tokens), output_dim=1024))

In case, you haven't added input_length and input_shape parameters in the embedding layer, then 

input_shape = (None,)

else

input_shape = (input_length,) # added 'input_length=' parameter

For more info check the official code here.

Upvotes: 1

Related Questions