How often is the training and validation data updated?

Hi guys, I’ve just joined Numerai couple of days ago and I found that the training and validation data of round 252 and 253 are the same. Does anyone know how often the training and validation data are updated please?

1 Like

The training data never updates. That’s it. However, we do get a handful of new validation eras once in a while (once or twice a year maybe). They said they were gonna add some every six months, but it doesn’t look like they are actually gonna hold to that. There is a feature expansion in the pipeline though – i.e. not more training rows, but more features per row. A whole lot more (10x what we have now). There will be a big announcement about that when the time comes…


Thanks a lot! I’ve thought the data is constantly updated and worried that we need to re-train the model once a while…wondering how can the data keep the information up to date if it’s never updated…looking forward to the big announcement!

The training data represents 10 years, but the newest of it is at least a several years old. (There is also another 2 1/2 years worth of validation data, some of which is pretty recent.) The idea is to make a general model that works well…forever, I guess, and not just fit (overfit) to the most recent time periods. Whether or not that is a wise setup is open to debate, but that’s the way it is.