Random Forests for interpreting tICs

tic_rfr.py

from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
import numpy as np 

def compute_random_forests(features, projected_tica_coords, rf_dir, n_trees=500, 
  n_tica_components=25, start_tIC=0):

  features = np.concatenate(features)
  tics = np.concatenate(projected_tica_coords)

  for j in range(start_tIC, np.shape(tics)[1]):
    #os.system("rm %s/tIC.%d_rfr.pkl" %(rf_dir, j))
    rfr = RandomForestRegressor(n_estimators=n_trees, max_features="sqrt",
                                n_jobs=-1, verbose=1)
    print(("Fitting tree for tIC %d" %(j+1)))
    rfr.fit(features, tics[:,j])
    print("Done fitting RF model. Saving now...")

    with open(os.path.join(rf_dir, "tIC.%d_rfr.pkl" %j), "wb") as f:
      pickle.dump(rfr.feature_importances_, f)
    print("Saved random forest feature importances")

  return