Skip to content

Latest commit

 

History

History
249 lines (175 loc) · 11.4 KB

CONTRIBUTING.md

File metadata and controls

249 lines (175 loc) · 11.4 KB

Contributing

Contributor license agreement

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

Before contributing, please check that you have the correct configurations of (1) Contributor License Agreement and (2) GPG key.

(1) Contributor License Agreement When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

(2) GPG Key

Setting up a GPG key has three stages:

  1. Generate the key
  2. Tell GitHub about the key
  3. Instruct Git to sign using your key

Note that the GitBash shell installed by Git on Windows already has GPG installed, so there is no need to install GPG separately.

Please also make sure to set email in git config (git config --global user.email "MY_NAME@example.com") which should be the same email linked to the PGP Key.

If you have previously committed changes that were not signed follow these steps to sign them retroactively after setting up your GPG key as described in the GitHub documentation.

Code of conduct

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Acceptance criteria

All pull requests need to abide by the following criteria to be accepted:

  • passing pipelines on the GitHub pull request
  • signed Contributor License Agreement (CLA)
  • approval from at least one maintainer
  • compatibility with light / dark / high-contrast themes
  • fits with overall look-and-feel of the widget
  • accessibility (to be clarified)
  • support for localization in code (translations need not be provided)
  • tests for added / changed functionality

Development process

First ensure you have npm installed (which in turn may require installing node).

This repository is tested with node 14.x, 15.x and 16.x, so using the latest 16.x version is recommended.

Using npm you can install yarn as follows:

npm install -g yarn

If yarn --version succeeds you can proceed. If not, you may have to follow the instructions printed by your shell. If you're using Powershell you may have to bypass the execution policy to allow yarn to execute. One way to do this is Set-ExecutionPolicy -ExecutionPolicy Bypass.

For all further steps yarn install is a prerequisite. Run the yarn install command from your repository root directory.

To run the dashboards locally run the following from the root of the repository on your machine:

yarn start

which can take a few seconds before printing out

$ nx serve

> nx run dashboard:serve
**
Web Development Server is listening at http://localhost:4200/
**

at which point you can follow the link to your browser and select the dashboard and version of your choice.

Linting

To check for linting issues and auto-apply fixes where possible run

yarn lintfix

You could also use prettier to lint your files. Once you have installed prettier using yarn, you could use yarn prettier --write . to lint files in a particular directory.

Building

To build a specific app run

yarn build <app-name>  // e.g. fairness, interpret

or alternatively yarn buildall to build all of them. Since most apps have dependencies on mlchartlib it makes sense to run yarn buildall at least once.

Testing

Run e2e tests locally with mock data

  1. git clone https://github.com/microsoft/responsible-ai-toolbox
  2. cd responsible-ai-toolbox
  3. yarn install
  4. yarn build
  5. To execute tests run yarn e2e. Sometimes it is preferable to watch the execution and select only individual test cases. This is possible using yarn e2e --watch.

cypress window will open locally - select test file to run the tests

Run e2e tests locally with notebook data

  1. git clone https://github.com/microsoft/responsible-ai-toolbox

  2. cd responsible-ai-toolbox (It is recommended to create a new virtual environment and install the dependencies)

  3. yarn install

  4. yarn buildall or yarn build widget

  5. pip install -e responsibleai_vision if using the RAI Vision Dashboard locally.

  6. pip install -e responsibleai_text if using the RAI Text Dashboard locally.

  7. pip install -e raiwidgets to install raiwidgets locally.

  8. pip install -e responsibleai to install responsibleai locally.

    If there are changes to other python packages, you will want to install them locally as well:

    pip install -e raiutils

    pip install -e erroranalysis

    pip install -e rai_core_flask

    pip install -e nlp_feature_extractors if using the RAI Text Dashboard locally.

    If there are no changes to them, then raiwidgets install will pick up the latest versions released on pypi.

  9. pip install jupyter

  10. cd notebooks\responsibleaidashboard

  11. To execute tests run yarn e2e-widget. Sometimes it is preferable to watch the execution and select only individual test cases. This is possible by running the notebook manually and using yarn e2e-widget -w --host {host} -n {notebook} where host is where RAI widget runs on (printed in notebook output) and notebook is the name of the notebook you are running. Eg: yarn e2e-widget -w --host 8704 -n responsibleaidashboard-census-classification-model-debugging

Cypress window will open locally - select test file to run the tests.

Since it may take a while to generate and execute all notebooks which makes the interactive --watch mode tedious, there's an option -n to specify individual notebooks. The argument is the notebook name without path.

Example: -n responsibleaidashboard-diabetes-regression-model-debugging

Currently, only a single notebook can be specified.

The notebooks can also be run with flights enabled. For that, simply add your preferred flights with -f.

Example with a single flight flightName: -f flightName

Example with multiple flights f1 and f2: -n f1,f2

Furthermore, when iterating on writing such tests it may not be necessary to regenerate the notebook(s) every single time. To avoid wasting time on this there's an option --skipgen to skip the notebook generation.

Test UX and SDK changes

For any new change, which involves changing any of the python SDK components and UI components, the manual testing of the code change can be done using the following steps:

  1. git clone https://github.com/microsoft/responsible-ai-toolbox
  2. cd responsible-ai-toolbox (It is recommended to create a new virtual environment and install the dependencies)
  3. You should commit all your current set of changes for SDK and UX using git commit.
  4. Clean all untracked files using git clean -fdx
  5. Run yarn install and yarn buildall to build the UX changes.
  6. Run pip install -e responsibleai_vision if using the RAI Vision Dashboard locally.
  7. Run pip install -e responsibleai_text if using the RAI Text Dashboard locally.
  8. Run pip install -e raiwidgets to install raiwidgets locally.
  9. Run pip install -e responsibleai to install responsibleai locally.
  10. Run pip install -e erroranalysis to install erroranalysis locally.
  11. Run pip install -e rai_core_flask to install rai_core_flask locally.
  12. Run pip install -e raiutils to install raiutils locally.
  13. Run pip install -e nlp_feature_extractors to install nlp_feature_extractors locally if using the RAI Text Dashboard.
  14. Run pip install -e rai_test_utils to install rai_test_utils locally.
  15. Install jupyter using pip install jupyter
  16. Open any notebook using python SDK and any widget from responsible-ai-toolbox and test your changes.

The steps from 3 to 14 need to be repeated if you incrementally change UI or SDK.

Debugging

There are several different ways to debug the dashboards:

  1. Use Chrome + React Developer Tools. The debugging experience can be a bit flaky at times, but when it works it allows you to set breakpoints and check all variables at runtime.

  2. Adding console.log(...) statements and check the console during execution. Please remember to remove the statements later on.

  3. Alternatively, you can set objects as part of window to inspect them through the console at runtime (as opposed to having to specify what to print with console.log at compile time).

Flighting

It is possible to create feature flights to use certain functionality under development before exposing it to all users immediately. To do so, go to responsible-ai-toolbox\libs\model-assessment\src\lib\ModelAssessmentDashboard\FeatureFlights.ts and add your flight. After that you can use it in Typescript code as follows:

isFlightActive(flightName, this.context.featureFlights);

To pass the flight into the ResponsibleAIDashboard, simply add the keyword argument feature_flights and separate all the flights you wish to pass with ampersand (&), e.g., feature_flights="flight1&flight2&flight3".

In the dashboard test environment (using yarn start) you have a dropdown to select which flights should be active.

Code approvals

Once you have made your code changes locally, committed them and verified them, you can send a pull request (in short form written as PR) to responsible-ai-toolbox. For more information on how to create a pull request, please see Proposing changes to your work with pull requests.

The PR will need to be approved by at least one code reviewer. In addition, if any changes are made in the listed directories within the code owners file, those owners will be required to approve the PR. You can tag those owners directly in the comments to ensure they are aware of the changes made. Only one code owner is required for an area, but if the PR makes changes in multiple areas at least one code reviewer will be required from each area, hence multiple code reviewers could be required. In general, it is better to make more smaller PRs than fewer larger PRs to make it easier to review the code. Please ensure all automated builds/tests pass on the PR.