Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://fosstodon.org/users/ataustin/statuses/111093662357195553">Adam @ Posit Conf (ataustin@fosstodon.org)'s status on Wednesday, 20-Sep-2023 21:47:26 JST</a><a href="https://fosstodon.org/@ataustin" title="ataustin@fosstodon.org"><img src="https://gnusocial.jp/avatar/174553-48-20230920124726.webp" width="48" height="48" alt="Adam @ Posit Conf" style="position: absolute; left: 0; top: 0;">Adam @ Posit Conf</a><div><ul><li></ul></div></section><article><p>Maintaining ML models is like doing laundry -- it never ends! <a href="https://fosstodon.org/@juliasilge">@juliasilge</a> with the harsh realities of data science life.</p><p>ML has both software AND statistical properties. Failures in statistical performance can be *silent*.</p><p>Data drift (inputs changing) and concept drift (feature vs outcome relationships changing) are sneaky sources of changes over time.</p><p>Watch out - deploying a model may cause drift via feedback loops!</p><p>The slides are so good: <a href="https://juliasilge.github.io/ml-maintenance-2023/#/title-slide" rel="nofollow noreferrer">https://juliasilge.github.io/ml-maintenance-2023/#/title-slide</a></p><p><a href="https://fosstodon.org/tags/PositConf2023" rel="tag">#PositConf2023</a></p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/2081166#notice-4086415">In conversation</a><time datetime="2023-09-20T21:47:26+09:00" title="Wednesday, 20-Sep-2023 21:47:26 JST">Wednesday, 20-Sep-2023 21:47:26 JST</time> <span>from <span><a href="https://fosstodon.org/@ataustin/111093662357195553" rel="external" title="Sent from fosstodon.org via ActivityPub">fosstodon.org</a></span></span><a href="https://fosstodon.org/@ataustin/111093662357195553">permalink</a><h4>Attachments</h4><ol><li><label><a rel="external" href="https://gnusocial.jp/attachment/1545665">Untitled attachment</a></label><br><a href="https://cdn.fosstodon.org/media_attachments/files/111/093/656/643/494/243/original/3f7963f5b11a3119.png" rel="external">https://cdn.fosstodon.org/media_attachments/files/111/093/656/643/494/243/original/3f7963f5b11a3119.png</a></li><li><article><header><div>Domain not in remote thumbnail source whitelist: cdn2.dan.com</div><h5><a href="https://life.ml/">life.ml - Domain Name For Sale | Dan.com</a></h5><div> from <span>@undeveloped</span></div></header><div>I found a great domain name for sale on Dan.com. Check it out!</div><footer></footer></article></li><li><article><header><div>No result found on File_thumbnail lookup.</div><h5><a href="https://juliasilge.github.io/ml-maintenance-2023/#/title-slide#PositConf2023">Reliable maintenance of machine learning models</a></h5><div></div></header><div></div><footer></footer></article></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
Adam @ Posit Conf (ataustin@fosstodon.org)'s status on Wednesday, 20-Sep-2023 21:47:26 JSTAdam @ Posit Conf
- Julia Silge
Maintaining ML models is like doing laundry -- it never ends! @juliasilge with the harsh realities of data science life.
ML has both software AND statistical properties. Failures in statistical performance can be *silent*.
Data drift (inputs changing) and concept drift (feature vs outcome relationships changing) are sneaky sources of changes over time.
Watch out - deploying a model may cause drift via feedback loops!
The slides are so good: https://juliasilge.github.io/ml-maintenance-2023/#/title-slide
#PositConf2023
In conversationWednesday, 20-Sep-2023 21:47:26 JST from fosstodon.orgpermalink
Attachments
1. Untitled attachment
  https://cdn.fosstodon.org/media_attachments/files/111/093/656/643/494/243/original/3f7963f5b11a3119.png
2. Domain not in remote thumbnail source whitelist: cdn2.dan.com
  life.ml - Domain Name For Sale | Dan.com
  from @undeveloped
  I found a great domain name for sale on Dan.com. Check it out!
3. No result found on File_thumbnail lookup.
  Reliable maintenance of machine learning models

Public

Embed Notice

HTML Code

Corresponding Notice