Numerai Signals Roundtable II, hosted by Richard Craib and Liam Hinzman

Timestamps:
[0:08] Surajp made an example script that works with quandl. It is currently being tested and will be included when you download example scripts. [coming soon]

[0:38] Carlo Lepelaars’ sentiment script possibly being added

[01:02] objectscience’s script - a more evolved version of jrAI’s script.
This could be interesting as a patch to jrAI’s script in the future. [link coming soon]

[03:23] We are also looking into improving some of the other data that we give to people, such as the public ticker map. Currently working on the Bloomberg ticker and Yahoo ticker to make it update automatically every week.

[4:18] On the issue of Taiwanese and Canadian ticker corrections

[04:40] There is a community ticker map, built off of our public ticker map, with Alpha Vantage tickers being added.

[05:35] Bloomberg ticker with exchange code would be preferable to composite code

[07:37] Where should the community go by default, the Numerai ticker map or the community ticker map?

[11:44] People are requesting graphs for their validation diagnostics, what should we keep in mind when implementing this?

[13:12] The current agenda for Signals updates and improvements

Slido questions and feature requests:
[18:25] Is there a way to get a list of your invalid tickers, instead of just a number?

[19:21] Are the features used in Signals(for FN corr) and Classic(310) tournament shared? Will Signals corr be affected after 3100 features in classic?

[26:20] Would you please provide our per-era validation correlations after submission? Yes, that will be happening in the future.

[26:32] If your predictions are neutralised to the main tournament features, would a model using the main tournament features with positive fnc also do well in signals?

[30:38] Is Bloomberg your price provider for calculating the validation targets and scoring live submissions?

[37:15] Do you think a complicated ML prediction is strictly better than a simple measure?

[41:21] Request: ADD simpler more transparent separate tournaments to drive adoption cos opaque + hard data barrier (e.g. US stocks only, raw returns, Fama 3-factor alpha as target)

[44:56] Could you brainstorm alternative datasets ideas?

[47:12] Would you please change from Bloomberg Composite to Bloomberg Exchange ticker? Korean stocks are very difficult to match without exchange code? Discussed earlier, at 05:35.

[47:20] Would you consider scoring on a smaller subset of the universe since obtaining data for the full universe is extremely difficult, even impossible?

[49:11] Having a Numerai tournament-like structure where features are provided will remove barrier to entry. Mention & source free data (3-5 features), rather than users bringing their own features.
--
Numerai Signals is Numerai's new tournament where you can create your signal with any dataset and model. Submit on any of the ~5000 global stocks in Numerai’s universe. Your signal is compared with other signals so you can discover how original it is. Original signals earn the most.
--
Are you a data scientist without your own data?
Download our high quality hedge fund data, which we've obfuscated so that we can give it to you for free. It is clean and regularized, designed to be usable right away.