Keras masking layer performance

py

import keras
from keras.layers import *
from keras.models import Model
import theano as T
import tensorflow as tf
print('theano ver.',T.__version__) 
print('tensorflow ver.',tf.__version__) 
print('keras ver.',keras.__version__) 
np.set_printoptions(precision=4)
np.random.seed(1)

from keras.models import Sequential

import timeit
def lstm_masked():

    model = Sequential()
    model.add(Masking(mask_value = mask_value,input_shape=(n_timesteps, n_features)))
    model.add(LSTM(20, return_sequences=True))
    model.add(LSTM(20, return_sequences=True))

    output_val = model.predict(data)
    return output_val
    
def lstm_nonmasked():
    model = Sequential()
    model.add(LSTM(20, return_sequences=True, input_shape=(n_timesteps, n_features)))
    model.add(LSTM(20, return_sequences=True))

    output_val = model.predict(data)
    return output_val

n_samples  = 1000
n_timesteps = 1000
n_features = 20
mask_value = -1

data = np.random.uniform(0,1,(n_samples,n_timesteps,n_features))
data[:,:(n_timesteps/2),:] = mask_value

print('Time  without: ',timeit.timeit(lstm_nonmasked, number=2))
print('Time  with   : ',timeit.timeit(lstm_masked, number=2))
data = np.random.uniform(0,1,(n_samples,n_timesteps,n_features))
data[:,:(n_timesteps/2),:] = mask_value
print('Time  with   : ',timeit.timeit(lstm_masked, number=2))
print('Time  without: ',timeit.timeit(lstm_nonmasked, number=2))

# TENSORFLOW BACKEND
# ('tensorflow ver.', '1.0.1')
# ('keras ver.', '2.0.1')
# ('Time  without: ', 90.77176785469055)
# ('Time  with   : ', 89.05559015274048)
# ('Time  with   : ', 91.29713892936707)
# ('Time  without: ', 90.39765214920044)
# KERAS BACKEND
# ('theano ver.', '0.8.2')
# ('Time  without: ', 16.689538955688477)
# ('Time  with   : ', 20.254252910614014)
# ('Time  with   : ', 15.486266136169434)
# ('Time  without: ', 13.057673215866089)
# Can someone scale this up?