Reproducibility of treatment effects with random forest and honest trees in final model

[Chandrak1907]

I appreciate your great work. I am trying to use random forest and honest trees to estimate treatment effect by specifying 'random_state' argument. However, I get different results in each iteration.

Working example with data given in ''Double Machine Learning Examples.ipynb' is here.

And for my specific use case, results differ a lot in each iteration. So, I am curious to know how to trust the treatment effect results.

[kbattocchi]

Thanks for the very clear bug report, we'll take a look.

[kbattocchi]

This should be addressed by #258. We're planning to put out a release including these changes next week.

[Chandrak1907]

models can be saved now.