Keras error : Expected to see 1 array

Question

I got the following error when I tried to train an MLP model in keras(I am using keras version 1.2.2)

Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 1 arrays but instead got the following list of 12859 arrays:

This is the summary of the model

____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to
====================================================================================================
dense_1 (Dense)                  (None, 20)            4020        dense_input_1[0][0]
____________________________________________________________________________________________________
dense_2 (Dense)                  (None, 2)             42          dense_1[0][0]
====================================================================================================
Total params: 4,062
Trainable params: 4,062
Non-trainable params: 0
____________________________________________________________________________________________________
None

This is the first line of model

 model.add(Dense(20, input_shape=(200,), init='lecun_uniform', activation='tanh'))

For training:

model.fit(X,Y,nb_epoch=100,verbose=1)

where X is a list of elements and each element in turn is a list of 200 values.

Edit :

I also tried

model.add(Dense(20, input_shape=(12859,200), init='lecun_uniform', activation='tanh'))

but I am getting the same error

Solution 1

Your error comes from the fact that your X for some reason wasn't transformed to a numpy.array. In this your X is treated as a list of rows and this is a reason behind your error message (that it expected one input instead of list which has a number of rows elements). Transformation:

X = numpy.array(X)
Y = numpy.array(Y)

I would check a data loading process because something might go wrong there.

UPDATE:

As it was mentioned in a comment - input_shape need to be changed to input_dim.

UPDATE 2:

In order to keep input_shape one should change to it to input_shape=(200,).

Solution 2

I fixed mine by adding

np.array

to train_X , train_Y , valid_X and valid_Y. For example,

model.fit(np.array(train_X),np.array(train_Y),
          batch_size=32,nb_epoch=20,
          validation_data=(np.array(valid_X),np.array(valid_Y)),
          callbacks=[early_stop])

I got the help from here. This approach is likely to have a slow run because all data features will have to be converted to numpy arrays and it could be a lot of work for your system.