Reproducible results using Keras with TensorFlow backend

Question

I am using Keras to build a deep learning LSTM model, using TensorFlow backend. Each time I run the model, the result is different. Is there a way to fix the seed to create

Answer 1

Basically, the key idea of making the result reproducible is to disable GPU. This is very important. To do this, just include

import os
import tensorflow as tf
import numpy as np
import random as rn

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = ""

sd = 1 # Here sd means seed.
np.random.seed(sd)
rn.seed(sd)
os.environ['PYTHONHASHSEED']=str(sd)

from keras import backend as K
config = tf.ConfigProto(intra_op_parallelism_threads=1,inter_op_parallelism_threads=1)
tf.set_random_seed(sd)
sess = tf.Session(graph=tf.get_default_graph(), config=config)
K.set_session(sess)

at the very beginning your code. Hope this can help you.