Is possible download last tournaments datasets?


Hi, everyone, I’m a IA studen and i’m interestet to work with numeria with academical intencions.
I only I want to know (for the moment) if there are avaliable last tournaments’s datastets and not only currents. And if I can download they.
Tanks you.


The dataset has been the same for the last 8 rounds I think, so if you have the current dataset you should be good.

The current dataset had 50 features, before this format, the dataset had 20 features. So even if you could find the older datasets, they would not be useful at all.


Do you mean, no change whatsoever – in the ENTIRE dataset? Every id and feature, in every row (of training, validation, and live), has not changed over 8 rounds?

I thought that the “live” part of the dataset, was what Numerai (the hedge fund) used to steer their week-to-week trading.


The training data has not changed. The tournament data file (which contains the validation and live data) has changed. But that data should not be used for training purposes anyway. If you train your models using the validation data you are likely overfitting and getting artificially low logloss values on the leaderboard.


Thanks for the reply (back in 2017/11); I’ve been thinking about the possible consequences, of not changing the training data, and will shortly express those in a new post.