CrossTrainer: Practical Domain Adaptation with Loss Reweighting

This is an implementation of the method described in "CrossTrainer: Practical Domain Adaptation with Loss Reweighting" by Justin Chen, Edward Gan, Kexin Rong, Sahaana Suri, and Peter Bailis.

Install

The crosstrainer package can be installed using pip.

pip install crosstrainer

Usage

CrossTrainer utilizes loss reweighting to train machine learning models using data from a target task with supplementary source data.

Inputs:

Base model, target data, source data.

Outputs:

Trained model with optimized weighting parameter alpha.

Example:

from crosstrainer import CrossTrainer
from sklearn import linear_model

lr = linear_model.LogisticRegression()
ct = CrossTrainer(lr, k=5, delta=0.01)
lr, alpha = ct.fit(X_target, y_target, X_source, y_source)
y_pred = lr.predict(X_test)

More examples can be found in the tests file: crosstrainer/tests/test_crosstrainer.py.