2

I'm fitting TPOTClassifier and it currently takes considerable time. Is it any possibility to store all the "population" and all the settings needed to have them loaded after and continue training further? What I read in the documentation allows to store only the best "evolved" pipeline using export method:

tpot = TPOTClassifier(generations=5, population_size=50, verbosity=2)
tpot.fit(X_train, y_train)
print(tpot.score(X_test, y_test))
tpot.export('tpot_best_pipeline.py') # export the best pipeline among the population

Also I can limit a number of generations by passing generations argument in TPOTClassifier constructor as above, but how can I persist the classifier?

stop-cran
  • 4,229
  • 2
  • 30
  • 47
  • 3
    Have you tried to pickle the TPOTClassifier instance? Especially when combined with the `warm_start` feature, you should be able to pause the run partway through, pickle it, then load it back via pickle and pick up where you left off. – Randy Olson Mar 02 '18 at 18:57

0 Answers0