fmohr / asforests

Automatically Stopping Random Forests - a sklearn extension

asforests

Automatically Stopping Random Forests - an sklearn extension

Why to use?

Never worry again about the number of trees you should use in the forest: Standard Random Forest implementations like in sklearn require a hyperparameter for the number of trees. However, this number can be determined automatically online analyzing the performance curve of the forest. This is what asforests do.

Implementations in asforests inherit from the forests in scikit-learn, so they can be used exactly in the same way, except that you do not (cannot) specify the number of trees.

Usage

Installation

pip install asforests

Example Code

from asforests import RandomForestClassifier
rf = RandomForestClassifier()

To test this with some dataset:

import sklearn.model_selection
import sklearn.datasets
X, y = sklearn.datasets.load_iris(return_X_y = True)
sklearn.model_selection.cross_validate(rf, X, y)

Remarks

• The RandomForestClassifier in asforests supports all the hyperparameters of the standard RandomForestClassifier from scikit-learn, with the same sematics.
• Currently, the forest internally uses a validation fold to build the learning curve. This can imply that the performance of the RF is slightly worse that the one of the standard RF, because less instances are used for training. This will be replaced in a future version by the OOB estimate.

asforests's Issues

use OOB score instead of a validation score

to prevent that the performance is worse than the one of a standard RF.

enable support for pandas dataframes