Extracting Fitted Honest Forest from CausalForest or DROrthoForest?

Thanks so much for this package! It is great to have all these methods together in a comprehensive library.

I am working on a project and want to calculate a covariate balance test using the honest forest, similar to Figure 2 in this paper (https://arxiv.org/abs/1909.09138). I have also included a screenshot of Figure 2 at the bottom, where the blue dots are the raw differences between treatment and control, the green dots are the differences using matching, and the pink dots are from causal forest matching. In order to do this, I need to first train the CausalForest or DROrthoForest. Then I replace the outcome with each of the pre-treatment features and calculate the 'treatment effect' on the pre-treatment feature.

Do you know how to extract the trained forest? Specifically, here is simulation code based on the example from the documentation site:

import numpy as np
import sklearn
from econml.ortho_forest import ContinuousTreatmentOrthoForest, DROrthoForest
from econml.causal_forest import CausalForest

np.random.seed(123)

T = np.array([0, 1]*60)
W = np.array([0, 1, 2, 0]*30).reshape(-1, 1)
Y = (.2 * W[:, 0] + np.random.uniform(-2,2, len(T))) * T + .5 + np.random.uniform(-1,1, len(T) )

est = CausalForest(n_trees=5, max_depth=2, subsample_ratio=0.5,
                                   model_T=sklearn.linear_model.LogisticRegression(),
                                   model_Y=sklearn.linear_model.LinearRegression())

### What do I call next to recover the trained causalforest?

Thanks so much!

Start with the CausalForest, DROrthoForest, and ContinuousTreatmentOrthoForest entry points shown in the issue and review the documentation example around model training. Determine whether the trained forest can be extracted through the existing estimator interface; done should be a documented supported approach or a clearly stated limitation.