Functions to estimate importance of features in determining predictions for individual samples (aka "feature activations"). Fast nogil implementation in Cython.

import numpy as np
from sklearn.ensemble import RandomForestRegressor
from DTAnalyze.Activation import GetActivations

A = np.random.rand(256, 3)
Y = (2 * (A[:, 0] > 0.5) - (A[:, 1] < 0.5) - 
     (A[:, 2] > 0.5) + np.random.normal(0, 0.1, size=256))

rfr = RandomForestRegressor(n_jobs=4).fit(A, Y)

L1 = GetActivations(rfr, A)

Install using pip:

pip install DTAnalyze

Or you can build manually from this repository:

python setup.py build_ext

Then copy build artifact into DTAnalyze (sub) folder and put that folder somewhere in your path.