How to get reproducible results in keras

You can find the answer at the Keras docs: https://keras.io/getting-started/faq/#how-can-i-obtain-reproducible-results-using-keras-during-development.

In short, to be absolutely sure that you will get reproducible results with your python script on one computer's/laptop's CPU then you will have to do the following:

1. Set the PYTHONHASHSEED environment variable at a fixed value
2. Set the python built-in pseudo-random generator at a fixed value
3. Set the numpy pseudo-random generator at a fixed value
4. Set the tensorflow pseudo-random generator at a fixed value
5. Configure a new global tensorflow session

Following the Keras link at the top, the source code I am using is the following:

# Seed value
# Apparently you may use different seed values at each stage
seed_value= 0

# 1. Set the `PYTHONHASHSEED` environment variable at a fixed value
import os
os.environ['PYTHONHASHSEED']=str(seed_value)

# 2. Set the `python` built-in pseudo-random generator at a fixed value
import random
random.seed(seed_value)

# 3. Set the `numpy` pseudo-random generator at a fixed value
import numpy as np
np.random.seed(seed_value)

# 4. Set the `tensorflow` pseudo-random generator at a fixed value
import tensorflow as tf
tf.random.set_seed(seed_value)
# for later versions: 
# tf.compat.v1.set_random_seed(seed_value)

# 5. Configure a new global `tensorflow` session
from keras import backend as K
session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
sess = tf.Session(graph=tf.get_default_graph(), config=session_conf)
K.set_session(sess)
# for later versions:
# session_conf = tf.compat.v1.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
# sess = tf.compat.v1.Session(graph=tf.compat.v1.get_default_graph(), config=session_conf)
# tf.compat.v1.keras.backend.set_session(sess)

It is needless to say that you do not have to to specify any seed or random_state at the numpy, scikit-learn or tensorflow/keras functions that you are using in your python script exactly because with the source code above we set globally their pseudo-random generators at a fixed value.

Theano's documentation talks about the difficulties of seeding random variables and why they seed each graph instance with its own random number generator.

Sharing a random number generator between different {{{RandomOp}}} instances makes it difficult to producing the same stream regardless of other ops in graph, and to keep {{{RandomOps}}} isolated. Therefore, each {{{RandomOp}}} instance in a graph will have its very own random number generator. That random number generator is an input to the function. In typical usage, we will use the new features of function inputs ({{{value}}}, {{{update}}}) to pass and update the rng for each {{{RandomOp}}}. By passing RNGs as inputs, it is possible to use the normal methods of accessing function inputs to access each {{{RandomOp}}}’s rng. In this approach it there is no pre-existing mechanism to work with the combined random number state of an entire graph. So the proposal is to provide the missing functionality (the last three requirements) via auxiliary functions: {{{seed, getstate, setstate}}}.

They also provide examples on how to seed all the random number generators.

You can also seed all of the random variables allocated by a RandomStreams object by that object’s seed method. This seed will be used to seed a temporary random number generator, that will in turn generate seeds for each of the random variables.

>>> srng.seed(902340)  # seeds rv_u and rv_n with different seeds each

I finally got reproducible results with my code. It's a combination of answers I saw around the web. The first thing is doing what @alex says:

1. Set numpy.random.seed;
2. Use PYTHONHASHSEED=0 for Python 3.

Then you have to solve the issue noted by @user2805751 regarding cuDNN by calling your Keras code with the following additional THEANO_FLAGS:

3. dnn.conv.algo_bwd_filter=deterministic,dnn.conv.algo_bwd_data=deterministic

And finally, you have to patch your Theano installation as per this comment, which basically consists in:

4. replacing all calls to *_dev20 operator by its regular version in theano/sandbox/cuda/opt.py.

This should get you the same results for the same seed.

Note that there might be a slowdown. I saw a running time increase of about 10%.

The problem is now solved in Tensorflow 2.0 ! I had the same issue with TF 1.x (see If Keras results are not reproducible, what's the best practice for comparing models and choosing hyper parameters? ) but

import os
####*IMPORANT*: Have to do this line *before* importing tensorflow
os.environ['PYTHONHASHSEED']=str(1)

import tensorflow as tf
import tensorflow.keras as keras
import tensorflow.keras.layers 
import random
import pandas as pd
import numpy as np

def reset_random_seeds():
   os.environ['PYTHONHASHSEED']=str(1)
   tf.random.set_seed(1)
   np.random.seed(1)
   random.seed(1)

#make some random data
reset_random_seeds()
NUM_ROWS = 1000
NUM_FEATURES = 10
random_data = np.random.normal(size=(NUM_ROWS, NUM_FEATURES))
df = pd.DataFrame(data=random_data, columns=['x_' + str(ii) for ii in range(NUM_FEATURES)])
y = df.sum(axis=1) + np.random.normal(size=(NUM_ROWS))

def run(x, y):
    reset_random_seeds()

    model = keras.Sequential([
            keras.layers.Dense(40, input_dim=df.shape[1], activation='relu'),
            keras.layers.Dense(20, activation='relu'),
            keras.layers.Dense(10, activation='relu'),
            keras.layers.Dense(1, activation='linear')
        ])
    NUM_EPOCHS = 500
    model.compile(optimizer='adam', loss='mean_squared_error')
    model.fit(x, y, epochs=NUM_EPOCHS, verbose=0)
    predictions = model.predict(x).flatten()
    loss = model.evaluate(x,  y) #This prints out the loss by side-effect

#With Tensorflow 2.0 this is now reproducible! 
run(df, y)
run(df, y)
run(df, y)

In Tensorflow 2.0 you can set random seed like this :

import tensorflow as tf
tf.random.set_seed(221)


from tensorflow import keras
from tensorflow.keras import layers


model = keras.Sequential( [ 
layers.Dense(2,name = 'one'),
layers.Dense(3,activation = 'sigmoid', name = 'two'),
layers.Dense(2,name = 'three')])

x = tf.random.uniform((12,12))
model(x)