Skip to main content

Quick access to uncertainty and confidence of Keras networks.

Project description

Uncertainty-Wizard

WARNING This is a pre-release, published while setting up the CI. The first official release will be deployed in a couple of days.

This library provides simple and transparent uncertainty and confidence quantification for fast-forward tensorflow.keras models.

Import uncertainty_wizard as uwiz, and you will amongst other things be able to do the following:

  • Perform MC-Dropout using the Sequential API (model=uwiz.models.StochasticSequential(my_layers)) `
  • Perform MC-Dropout on pre-trained models (model=uwiz.models.stochastic_from_keras(your_model))
  • Train Deep Ensembles in a memory and time efficient way with our lazily loaded and easily parallelized LazyEnsembles. (model=uwiz.models.LazyEnsemble(path='model', num_models=20, default_num_processes=4)

Installation

It's as easy as pip install uncertainty-wizard

Requirements

  • tensorflow >= 2.2.0
  • python 3.6* / 3.7 / 3.8

Note that tensorflow 2.4 has just been released. We will test and create compatibility with uncertainty wizard in the next couple of weeks. Until then, please stick to tensorflow 2.3.x.

*python 3.6 requires to pip install dataclasses

Documentation

A link to our documentation and user guide will be added here soon.

Examples

Our docs contain a list of jupyter based examples, which you can run in colab right away. You can find them here: (Link will be added soon)

Authors and Paper

Uncertainty Wizard was developed at the Università della Svizzera Italiana (USI) in Lugano by Michael Weiss under the supervision of Prof. Paolo Tonella. If you like uncertainty wizard and use it for research, you can cite us:

@inproceedings{Weiss2021,
  title={Fail-Safe Execution of Deep Learning based Systems through Uncertainty Monitoring},
  author={Weiss, Michael and Tonella, Paolo},
  booktitle={2021 IEEE 14th International Conference on Software Testing, 
    Validation and Verification (ICST)},
  year={2021},
  organization={IEEE},
  note={forthcoming}
}

A preprint and a tool paper which provides a deeper technical discussion of uncertainty_wizard will be added in January at latest.

Contributing

Issues and PRs are welcome! Before investing a lot of time for a PR, please open an issue first, describing your contribution. This way, we can make sure that the contribution fits well into this repository.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

uncertainty-wizard-0.0.20a2.tar.gz (35.1 kB view details)

Uploaded Source

Built Distribution

uncertainty_wizard-0.0.20a2-py3-none-any.whl (85.5 kB view details)

Uploaded Python 3

File details

Details for the file uncertainty-wizard-0.0.20a2.tar.gz.

File metadata

  • Download URL: uncertainty-wizard-0.0.20a2.tar.gz
  • Upload date:
  • Size: 35.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.2.0 pkginfo/1.5.0.1 requests/2.22.0 setuptools/45.2.0 requests-toolbelt/0.9.1 tqdm/4.48.2 CPython/3.8.5

File hashes

Hashes for uncertainty-wizard-0.0.20a2.tar.gz
Algorithm Hash digest
SHA256 507dfc83a967ad2c8260e9f524b123e53a457032baec10da2430cf7a64ca82a1
MD5 ff66264b7bb8f3761dd3abe4419cb1af
BLAKE2b-256 8c16b46f616a068fedac692f0744ef84de00b95207722eb687a2a2f6db7c2055

See more details on using hashes here.

File details

Details for the file uncertainty_wizard-0.0.20a2-py3-none-any.whl.

File metadata

  • Download URL: uncertainty_wizard-0.0.20a2-py3-none-any.whl
  • Upload date:
  • Size: 85.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.2.0 pkginfo/1.5.0.1 requests/2.22.0 setuptools/45.2.0 requests-toolbelt/0.9.1 tqdm/4.48.2 CPython/3.8.5

File hashes

Hashes for uncertainty_wizard-0.0.20a2-py3-none-any.whl
Algorithm Hash digest
SHA256 510a597a2392d0c6c95073d5b5a23d4ce52a52531a020006b56999518baf9ef7
MD5 1c0b34436504e8206bfd903efa15f6d9
BLAKE2b-256 7ce0aa2aa8a5a2ab755968e713b58c1347336bac8877774b5040b5a994c4b531

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page