How can I remove the randomness of a decision tree on sklearn?

Question

I'm trying to create a deterministic decision tree in python, but I've some issue.

If I run 2 times my script I get to different Decision Trees, because of the random_state of the algorithm.

I tried to fixed random_state( random_state=0) but It's still not working.

I'd like to remove the randomness of my decision tree but I can t find a clear solution.

You should check for more randomness in your code, like train_test_split that could be influencing the tree. — Stradivari, Jun 17 '19 at 12:44
Thanks for yours responses, - I checked in my code but the split between the train test split is fixed. - what do you mean by random.seed the generator ? — Corentin Moreau, Jun 17 '19 at 12:51

score 3 · Answer 1 · answered Jun 17 '19 at 17:36

3

Sklearn uses the random seed from numpy, so you can set the global seed at the start of your script with:

import numpy as np

np.random.seed(0)

answered Jun 17 '19 at 17:36

Stradivari

2,626
1
9
21

score 3 · Answer 2 · answered Jun 17 '19 at 20:38

The random_state argument should work but here are 2 different options

Option 1:

from sklearn.ensemble import RandomForestRegressor

rf = RandomForestRegressor(n_estimators=1000, criterion='mse', min_samples_leaf=4, 
                           random_state= 0)

This should return the same results every single time.

Scikit-learn does not use its own global random state; whenever a RandomState instance or an integer random seed is not provided as an argument, it relies on the numpy global random state, which can be set using numpy.random.seed

Option 2:

That being said, adding np.random.seed() before importing the RandomForestRegressor should also do the trick.

import numpy as np
np.random.seed(0)

Source: http://scikit-learn.org/stable/faq.html#how-do-i-set-a-random-state-for-an-entire-execution

How can I remove the randomness of a decision tree on sklearn?

2 Answers2