https://github.com/tensorflow/model-remediation

Model Remediation is a library that provides solutions for machine learning practitioners working to create and train models in a way that reduces or eliminates user harm resulting from underlying performance biases.
https://github.com/tensorflow/model-remediation

machine-learning tensorflow

Last synced: 10 days ago
JSON representation

Model Remediation is a library that provides solutions for machine learning practitioners working to create and train models in a way that reduces or eliminates user harm resulting from underlying performance biases.

• Host: GitHub
• URL: https://github.com/tensorflow/model-remediation
• Owner: tensorflow
• License: apache-2.0
• Archived: true
• Created: 2020-10-23T17:56:58.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2023-03-06T20:09:48.000Z (almost 2 years ago)
• Last Synced: 2025-01-16T14:26:15.838Z (14 days ago)
• Topics: machine-learning, tensorflow
• Language: Python
• Homepage: https://www.tensorflow.org/responsible_ai/model_remediation?hl=en
• Size: 1.44 MB
• Stars: 43
• Watchers: 15
• Forks: 20
• Open Issues: 7
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE
  • Code of conduct: CODE_OF_CONDUCT.md
  • Authors: AUTHORS

Awesome Lists containing this project

README

        
# TensorFlow Model Remediation

TensorFlow Model Remediation is a library that provides solutions for machine

learning practitioners working to create and train models in a way that reduces

or eliminates user harm resulting from underlying performance biases.

[![PyPI version](https://badge.fury.io/py/tensorflow-model-remediation.svg)](https://badge.fury.io/py/tensorflow-model-remediation)

[![Tutorial](https://img.shields.io/badge/doc-tutorial-blue.svg)](https://www.tensorflow.org/responsible_ai/model_remediation/min_diff/tutorials/min_diff_keras)

[![Overview](https://img.shields.io/badge/doc-overview-blue.svg)](https://www.tensorflow.org/responsible_ai/model_remediation)

## Installation

You can install the package from `pip`:

```shell

$ pip install tensorflow-model-remediation

```

Note: Make sure you are using TensorFlow 2.x.

## Documentation

This library contains a collection of machine learning remediation techniques

for addressing potential bias in a model.

Currently TensorFlow Model Remediation contains the below techniques:

*   MinDiff technique: Typically used to ensure that a model predicts the

    preferred label equally well for all values of a sensitive attribute.

    Helpful when trying to achieve [equality of

    opportunity](https://developers.google.com/machine-learning/glossary/fairness#equality-of-opportunity).

*   Counterfactual Logit Pairing technique: Typically used to ensure that a

    model’s prediction does not change between “counterfactual pairs”, where the

    sensitive attribute referenced in a feature is different. Helpful when

    trying to achieve

    [counterfactual fairness](https://developers.google.com/machine-learning/glossary/fairness#counterfactual-fairness).

We recommend starting with the

[overview guide](https://www.tensorflow.org/responsible_ai/model_remediation) to

get an idea of TensorFlow Model Remediation. Next try one of our interactive

guides like the

[MinDiff tutorial notebook](https://www.tensorflow.org/responsible_ai/model_remediation/min_diff/tutorials/min_diff_keras).

[Counterfactual tutorial notebook](https://www.tensorflow.org/responsible_ai/model_remediation/counterfactual/guide/counterfactual_keras).

```python

import tensorflow_model_remediation as tfmr

import tensorflow as tf

# Start by defining a Keras model.

original_model = ...

# Next pick the remediation technique you'd like to use. For example, a

# MinDiff implementation might look like the below:

# Set the MinDiff weight and choose a loss.

min_diff_loss = tfmr.min_diff.losses.MMDLoss()

min_diff_weight = 1.0  # Hyperparamater to be tuned.

# Create a MinDiff model.

min_diff_model = tfmr.min_diff.keras.MinDiffModel(

   original_model, min_diff_loss, min_diff_weight)

# Compile the MinDiff model as you normally would do with the original model.

min_diff_model.compile(...)

# Create a MinDiff Dataset and train the min_diff_model on it.

min_diff_model.fit(min_diff_dataset, ...)

```

#### *Disclaimers*

*If you're interested in learning more about responsible AI practices, including*

*fairness, please see Google AI's [Responsible AI Practices](https://ai.google/education/responsible-ai-practices).*

*`tensorflow/model_remediation` is Apache 2.0 licensed. See the

[`LICENSE`](LICENSE) file.*