How can i change Feature Importance so it isn't random anymore? in Python

user_15972838 0
Viewed 70

I am doing feature selection using feature importance which is based on a decision tree.

#Feature importance (source: https://towardsdatascience.com/feature-selection-techniques-in-machine-learning-with-python-f24e7da3f36e)
X = df.iloc[:,1:31]  #independent columns
y = df.iloc[:,0]    #target column = diagnosis

model = ExtraTreesClassifier()
model.fit(X,y)
print(model.feature_importances_) #use inbuilt class feature_importances of tree based classifiers

feat_importances = pd.Series(model.feature_importances_, index=X.columns)
feat_importances.nlargest(10).plot(kind='barh', color="lightskyblue")
plt.show()

But the results change everytime i start it again, probably somewhere there is a random number somewhere. Is there a way to set this once and then keep it the same so that the results don't change everytime the kernel is restarted?

Thank you.

UPDATE: I think I managed it using: model = ExtraTreesClassifier(random_state=1) is that the right way to do it?

feature-selection feature-extraction python
1 Answers

You can set a random_state parameter of ExtraTreesClassifier to some value. In this case you will get reproducible results. For example:

model = ExtraTreesClassifier(random_state=0)
user_96327740
Related