#era 'X' in live

1 messages · Page 1 of 1 (latest)

hearty arch
#

I've developed model that depends on numerical value of era. My problem is that in live data era is equal to 'X'.

  1. Is this because training and validation are updated weekly, but live data is updated weekly ?
  2. If last era in current validation data is e.g. 500, then era in live data is 505 (validation data for last era has targets from era 504 - 4 weeks in advance)? Or there is a bigger gap?
  3. Last question: if era can't be provided in live data but could be calculated by some method like in point 2, then how to incorporate it with pickle upload? As far I understood pickle function has only acess to live data.
steady kindle
summer moss
#

you can calculate the era on your own easily if you really want to. someone posted a function that does it on the forums.

steady kindle
summer moss
#

https://forum.numer.ai/t/numerai-rounds-to-era/5891

Numerai Forum
Numerai rounds to era

Hi, is there a mapping from the numerai rounds, like 339 resolved last week, to training/validation data eras? Round 1033 seems to be the latest era with main target not NaN. So I assume that round corresponds to 339? Rounds 1034 to 1038 inclusive all have NaN for the main target, so I presume they correspond to Numerai Rounds 344, 349, 354, 35...

#

i agree with @steady kindle that using era # as a feature is not a good idea. actually, i think I tried it a couple months ago for fun and got lower CORR on the validation set, but who knows, maybe it has TC 😉

prisma shell
#

How will get get the current round without access to the api though?

summer moss
#

??? u can calculate based on date

prisma shell
#

yeah thats true, assuming numerai never ever skips one. I like sources of truth I guess so that didnt even come to mind lol

steady kindle