CODEWITHSUNDEEP
pythonmachine-learningkerasvalueerror
Daniil Iaitskov
Daniil Iaitskov

Reputation: 6039

Feeding keras model with multiple inputs

I am trying to do a simple hello world with Keras and stuck. At the beginning I had 1 layer with 1 input and 1 output and it worked pretty well for a straight line approximation ;)

import numpy as np
from keras.models import Sequential
from keras.layers import Dense
from keras.optimizers import RMSprop
from keras.losses import mean_squared_error
mo = Sequential()
d = Dense(1, input_shape=(1,))
mo.add(d)
mo.summary()

mo.compile(loss=mean_squared_error, optimizer=RMSprop(lr=0.4), metrics=['accuracy'])

mo.trainable = True

for i in range(-100, 100):
    mo.train_on_batch(x = [i], y = [i])

After that I've got bravery for 2 input parameters:

d = Dense(1, input_shape=(2,))
for i in range(-100, 100):
    mo.train_on_batch(x = [np.array([i,i])], y = [i])

np.array([1,1]).shape # gives (2,)

Though I am getting an exception:

ValueError: Error when checking input: expected dense_53_input to have shape (2,) but got array with shape (1,)

I tried various combinations like [[i],[i]].

Upvotes: 0

Views: 1014

Answers (1)

today
today

Reputation: 33410

The first dimension is always the batch dimension in Keras. Batch size refers to the number of samples processed in a pass (forward and backward). When you specify the input_shape argument it does not include the batch dimension. Therefore, a network with input shape of (2,) takes input data of shape (?,2) where ? refers to the batch size. So you must pass arrays of shape (?,2):

mo.train_on_batch(x=[np.array([[i,i]])], y=[i])

since:

np.array([[i,i]]).shape   # it is (1,2)

Upvotes: 1

Related Questions