YAND: Yet Another Numerai Dashboard

[numerologist]

Ever wondered how you perform versus your fellow Numerati?
Benchmark models are too easy / not enough of embarrassment?
I present to you: yet another Numerai dashboard! Yep, that’s the name.

YAND: Yet Another Numerai Dashboard

Main features:

• First of its kind, apple-to-apple comparison of Numerai users with date range selection
• TC and CORRv2 summary table with filtering and multi-column sorting
• Simplicity: get all the info you need with no extra clicks or we’ll give you your money back

Credits and acknowledgments:

• @pschork for the favicon I stole from Discord memes
• Everyone who upvoted my discord post for the support
• @ia_ai and @jrai with their dashboards for inspiration
• @uuazed for numerapi, Numerai team for the API, plotly for plotly

Roadmap:

• Add a models page?
• Feedback and suggestions are welcome!

[wigglemuse]

So this is based on account-level data and not model-level data. So how is that aggregated/weighted?

[danzell]

Any idea on how account titles like grand master, master, etc. are created? Is this part of the api or did you do some sort of mapping?

[numerologist]

how is that aggregated/weighted?

Just an average of whatever you see in the charts. And the charts, in turn, come from user profiles, so it’s the average over cumulative stake-weighted-average CORRv2/TC of all models.

how account titles like grand master, master, etc. are created? Is this part of the API?

Yes. I think it’s part of the not-yet-released grandmasters proposal: Grandmaster Proposals - Google Docs
No real purpose here, just think of it as an easter egg, expert danzell

[numerologist]

A small update. I managed to get Numerai endpoints added to Allowlisted sites for free users: PythonAnywhere, so now community projects on pythonanywhere should be much easier to make. Heck, you can even try to do submissions from there

[numerologist]

Account-level MMC charts are now live on the dashboard.
Also, benchmark models now appear by default.

[numerologist]

The dashboard has been updated to reflect the current leaderboard sorting.

It’s interesting to see that some of the top participants are highly correlated… like, almost 100% correlated.

YAND.

[numerologist]

YAND 2.0 is out!

What’s new:

• Added a Models page
• Replaced TC chart with Payout metric (0.5xCORR 2xMMC)

Also, it looks like we’re pushing the limits of free hosting, so please be gentle

[numerologist]

A minor update: pushed some optimizations and caching, the website should be a little more robust now.

[eleven_sigma]

Really great! Thank you a lot!
For YAND 3.0 this is my ‘wish list’
Save groups of models and assign a label, like ‘top performers MMC’, ‘top stackers’, ‘My lgb models’,…
and then be able to select a group to watch, instead of one by one.

[numerologist]

YAND gained some new features!

YAND supports both user and model comparison for all 3 tournaments now: Main, Signals, and Crypto. This includes displaying the corresponding payout metrics for each of the tournaments. Check it out:
Signals:

Crypto:

YAND.

[numerologist]

I haven’t had much time for this project recently, so no UI changes, but for those who need it, I just pushed an updated version of YAND with the feature to support sets of models and users for all 3 tournaments.

To use it, simply add ?set=... in the URL, where ... is a comma-separated list of models/users for the appropriate dashboard.
For example,

• https://yand.pythonanywhere.com/main/?set=numerologist,svendaj,benchmark_models shows performance for the Main tournament for the specified 3 users only;
• https://yand.pythonanywhere.com/main/models?set=nmrology,super_model,numerai_swmm shows performance for the Main tournament for the specified 3 models only;
• https://yand.pythonanywhere.com/signals/models?set=shallow_alpha,simple_lgbm_ffn - for Signals models;
• https://yand.pythonanywhere.com/crypto/models?set=joe_the_degen_01,joe_the_degen_02,joe_the_degen_03,joe_the_degen_04 - for Crypto models etc.

Note: the URL param overrules whatever is set in the dropdown, so the dropdown will be ignored if the set param is present. You can still freely adjust the time range.

Using your browser bookmarks, you can achieve the groups @eleven_sigma requested, just with your browser instead of fancy UI.