import keras.backend as K

def cross-entropy(yTrue,yPred):
    return -K.sum(yTrue*K.log(yPred))

model.fit(input,input_matrix,epochs=1)
input.shape # shape=(nb_tweets,100)
# ex. tweet = "if you have" input = [11, 14, 21,  4, 26, 19,  7,  4, 26, 18, 19]

input_matrix.shape # shape=(nb_tweets,100,28) each character is encoded using one-hot encoding
# ex. a = (0,1,0,0,0,....0)