October 6, 2020

Yellowbrick Advisory Board Meeting Held on October 6, 2020 from 2030-2200 EST via Video Conference Call. Kristen McIntyre presided as Chair, Larry Gray serving as Secretary and Edwin Schmierer as Treasurer. Minutes taken by Larry Gray.

Attendees: Rebecca Bilbro, Benjamin Bengfort, Kristen McIntyre, Larry Gray, Prema Roman, Edwin Schmierer, Tony Ojeda and Adam Morris.


A broad overview of the topics for discussion in the order they were presented:

  1. Welcome (by Kristen McIntyre)

  2. Treasurer Update

  3. Summer 2020 Semester Retrospective

  4. Yellowbrick v1.2 Milestone planning

  5. Member Topics

    1. Departure from AppVeyor Usage

    2. Managing Dependency upgrades

    3. New Handling of Third-Party Contribs

  6. Additional focus for the Fall

Treasurer Update

There are no major issues to report. As a reminder, in January, we approved our budget for the year, which totaled $271.48. Since we have 9 advisors for this year, the dues totalled $30.17 per advisor. All dues have been paid and all members have been reimbursed for expenses incurred on behalf of YB.

2020 Annual Budget




Paid By







Read theDocs GoldMembership

$10/ month










Question: What, if any, budget changes do we anticipate in 2021? No Budget changes anticipated in 2021.

Summer 2020 Semester Retrospective

Here were some of the key achievements and successes this past semester?

Yellowbrick Statistics May - October:

Issues closed: 20

Issues from this time period that remain open: 6

PRs from this time period that remain open: 3

Approved PRs (Contributors): 19

Rebecca - 14 Big Props!, Ben - 1, @Tktran - 2, @AlderMartinez - 1, @Melonhead901 - 1

A Summary of Approved PR Topics - Complete changelog Since v1.1 can be found here. (https://github.com/DistrictDataLabs/yellowbrick/pull/1110)

Major events:

  1. Removed Appveyor from CI Matrix

  2. Split Prediction Error Plot from Residuals Plot

Minor events:

  1. Third-Party Estimator Wrappers/Checks

  2. Improved Cook’s Distance and Manifold Documentation

  3. Improved RankD Tests and Upgrade

  4. Reintegrated spec and verbose as pytest addopts

  5. Small fixes to Manual Alpha Selection and ROCAUC


  1. Matplotlib 3.3/3.2.2 Test/Image Failure

  2. Quick-Start Plot Directive Fix

  3. Fixed Errors/Warnings due to upgrading dependencies


What are some of the challenges and issues faced by the maintainer and contributor team this past semester (technical or otherwise)?

  • Covid-19: No one expected the global pandemic…


Who deserves special recognition for their contributions to Yellowbrick this past semester?

  • Adam for always doing such a great job with our Twitter account

  • Evergreen shoutout: to Ben for doing “Ben-type” things

  • Rebecca for being a PR Master Blaster


Ben proposed that we immediately release v1.2. The release would contain non-trivial changes. Votes were cast on whether we should issue a release and all in attendance voted that it be done immediately.

We also resolved that we would lock all dependency versions in place then proactively upgrade Yellowbrick in accordance to dependency changes (see Member Topics Below).

Member Topics

We discussed:

  1. Managing Dependency Upgrades (Discussion Leader: Ben)

    1. Overarching Question: How do we manage this technical debt?

      1. Different approaches proposed:

        1. The most active approach would be to deprecate old quickly and refactor our code to meet new upgrades. This would involve rewriting tests and Visualizers

        2. A less active approach would be to only handle depreciation warnings

        3. The weakest form of a passive approach would be to limit dependencies

        4. The strongest form of a passive approach would be to freeze the repo

      ii. Rebecca noted that determining which dependency changes cause the errors is the hardest problem we face and it takes hours of work to find them.

      1. She requests that we don’t add any new dependencies and that she will deny and PR that introduces them

      iii. Prema suggested that we convene on a case-by-case basis and discuss whether we should take an active/passive approach

      iv. Rebecca noted that PIP is going to be changing soon and we need to put it as a Topic for the next Meeting

      1. Ben proposed that we fix test dependency with == and maintain knowledge of upgrades. He notes that it is a risky approach because we have to stay on top of changes in dependencies.

    2. Decided ACTION

      1. We will freeze dependency before 1.2 release and be proactive about upgrading dependencies

  2. Departure from Appveyor usage. Due to trouble tracking down Errors. Rebecca deceived to abandon usage.

  3. Rebecca recommended a reading by Nadia Eghbal called Working in Public

  4. New handling of Third-Party Contribs (Discussion Leader Ben)

    1. The work was pursued because we are highly dependent on properties and attributes of Scikit-learn API and the way we handle inputs prevents third- party integration beyond scikit-learn. In response, we provided a new framework to handle third-party Estimators and move them to the contrib. We now provide a nice response when non-sklearn estimators don’t work.

  5. Focus for the Fall

    1. Ben suggested a blog on Skorch and Yellowbrick.

      1. It should represent the compatibility of Skorch and Yellowbrick

      ii. Rebecca recommends using her mini-lab on LSTM as a template

    2. Rebecca will be giving a talk that discusses some Yellowbrick at PyData Global Talk

Actions Items

  1. Release version 1.2

  2. Freeze dependency before 1.2 release