forestci: confidence intervals for scikit-learn forest algorithms

0.7 verified Fri May 01 auth: no python

forestci provides confidence intervals for random forest predictions using the infinitesimal jackknife method. It supports scikit-learn's RandomForestClassifier, RandomForestRegressor, ExtraTreesClassifier, and ExtraTreesRegressor. Version 0.7 includes bug fixes and improved documentation, with no breaking changes from 0.6. The library is stable but released infrequently.

pip install forestci

Common errors

error AttributeError: 'RandomForestRegressor' object has no attribute 'estimators_samples_' ↓

cause Scikit-learn version < 0.24 does not expose estimators_samples_ attribute used by forestci for inbag calculation.

fix

Upgrade scikit-learn to >= 0.24 or use a workaround: pass inbag matrix manually via inbag parameter or use a patch.

error ModuleNotFoundError: No module named 'forestci' ↓

cause forestci is not installed, or installed in a different environment.

fix

pip install forestci

error TypeError: random_forest_error() missing 1 required positional argument: 'X_test' ↓

cause Call signature is random_forest_error(forest, X_test, X_train) with both X_test and X_train required.

fix

Provide both X_test (data for which to compute errors) and X_train (training data).

Warnings

gotcha random_forest_error requires the training data (X_train) to compute the inbag matrix. If you pass the same data for prediction, ensure it is the training set. ↓

fix Use X_train as the 'inbag' argument or pass the training data explicitly.

gotcha calc_inference (formerly calc_corr) has been deprecated. Use random_forest_error for variance estimation and then compute confidence intervals manually. ↓

fix Replace calc_inference with random_forest_error( rf, X_test, X_train ) + manual CI computation.

gotcha In version 0.7, the random_forest_error function may raise AttributeError if the forest estimator doesn't have estimators_samples_ attribute (e.g., older sklearn versions). ↓

fix Upgrade scikit-learn to 0.24+ or use the forestci compatible version. Check estimator type.

Imports

calc_inference
```
from forestci import calc_inference
```
main function for computing confidence intervals
random_forest_error
```
from forestci import random_forest_error
```
computes bias-corrected error variance
forestci
wrong
```
from forestci import forestci
```
correct
```
import forestci
```
forestci is a module, not a function; do not import as a submodule

Quickstart

Compute confidence intervals for RandomForestRegressor predictions using the infinitesimal jackknife.

from sklearn.ensemble import RandomForestRegressor
from sklearn.datasets import make_regression
from forestci import random_forest_error
import numpy as np

X, y = make_regression(n_samples=100, n_features=4, noise=0.1, random_state=42)
rf = RandomForestRegressor(n_estimators=100, random_state=42)
rf.fit(X, y)
pred = rf.predict(X)
error_var = random_forest_error(rf, X, X)
# 95% confidence interval
ci = 1.96 * np.sqrt(error_var)
print(ci[:5])