The algorithm details are available at:
Sumanta Basu, Karl Kumbier, James B. Brown, Bin Yu, Iterative Random Forests to detect predictive and stable high-order interactions, PNAS https://www.pnas.org/content/115/8/1943
The implementation is a joint effort of several people in UC Berkeley. See the Authors.md for the complete list. The weighted random forest implementation is based on the random forest source code and API design from scikit-learn, details can be found in API design for machine learning software: experiences from the scikit-learn project, Buitinck et al., 2013.. The setup file is based on the setup file from skgarden.
Use conda to create a python3.6 environment. Run the following commands inside the package directory
pip install numpy cython==0.29 packaging
pip install -e .
(outdated) To install, simply run pip install irf
. If you run into any issues, see installation help.
In order to use irf, you need to import it in python.
import numpy as np
from irf import irf_utils
from irf.ensemble import RandomForestClassifierWithWeights
Generate a simple data set with 2 features: 1st feature is a noise feature that has no power in predicting the labels, the 2nd feature determines the label perfectly:
n_samples = 1000
n_features = 10
X_train = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_train = np.random.choice([0, 1], size=(n_samples,), p=[.5, .5])
X_test = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_test = np.random.choice([0, 1], size=(n_samples,), p=[.5, .5])
# The second feature (which is indexed by 1) is very important
X_train[:, 1] = X_train[:, 1] + y_train
X_test[:, 1] = X_test[:, 1] + y_test
Then run irf
all_rf_weights, all_K_iter_rf_data, \
all_rf_bootstrap_output, all_rit_bootstrap_output, \
stability_score = irf_utils.run_iRF(X_train=X_train,
X_test=X_test,
y_train=y_train,
y_test=y_test,
K=5, # number of iteration
rf = RandomForestClassifierWithWeights(n_estimators=20),
B=30,
random_state_classifier=2018, # random seed
propn_n_samples=.2,
bin_class_type=1,
M=20,
max_depth=5,
noisy_split=False,
num_splits=2,
n_estimators_bootstrap=5)
all_rf_weights stores all the weights for each iteration:
print(all_rf_weights['rf_weight5'])
The proposed feature combination and their scores:
print(stability_score)