prfr 0.1.6

No project description provided

prfr

Probabilistic random forest regressor: random forest model that accounts for errors in predictors and labels, yields calibrated probabilistic predictions, and corrects for bias.

Installation

pip install prfr

OR

pip install git+https://github.com/al-jshen/prfr

Example usage

import numpy as np
from prfr import ProbabilisticRandomForestRegressor, split_arrays

x_obs = np.random.uniform(0., 10., size=10000).reshape(-1, 1)
x_err = np.random.exponential(1., size=10000).reshape(-1, 1)
y_obs = np.random.normal(x_obs, x_err).reshape(-1, 1) * 2. + 1.
y_err = np.ones_like(y_obs)

train, test, valid = split_arrays(x_obs, y_obs, x_err, y_err, test_size=0.2, valid_size=0.2)

model = ProbabilisticRandomForestRegressor(n_estimators=250, n_jobs=-1)
model.fit(train[0], train[1], eX=train[2], eY=train[3])
model.calibrate(valid[0], valid[1], eX=valid[2], eY=valid[3], apply_bias=False)
model.fit_bias(valid[0], valid[1], eX=valid[2])

pred = model.predict(x_test, eX=x_err_test)
pred_qtls = np.quantile(pred, [0.16, 0.5, 0.84], axis=1)

print(pred.shape)