import numpy
from keras.models import Model
from keras.layers import Input, Dropout, TimeDistributed

batch_size = 10
timesteps = 5
nb_features = 7
dropout = 0.5

input = Input(batch_shape=(batch_size, timesteps, nb_features))
a = input
a = TimeDistributed(Dropout(dropout))(a) # gives error
output = a

model = Model(input, output)
model.compile("adam", "mse")

X = numpy.random.rand(batch_size, timesteps, nb_features)

Y = model.predict_on_batch(X)