Sorry if I’m not being clear, there are 120 eras in the training cohort, given that we have been told this is 10 years of data, it is reasonable to assume that each monthly era is equivalent to one month (out of 12, rather than 4 weeks out of 52). If this is the case (which i think it is), then the weekly eras will overlap the monthly eras in the test data (only by a bit).
so era 120 of training data is month 120. era 121 in val1 is month 121, through to era 212 in val2 which is month 212. test eras (which are in between val eras) are weekly so era 575 of test1 is actually a monthly era of 132 and then 576 is a monthly era of 132.5, 577 is monthly 132.8. and so on.
this is given the basis that there are 52.1429 weeks in a year, which when divided by 12 months = 4.34524. so divide each “weekly” era by this number to get the monthly era.
At least, this is the basis i have been working off and would love someone to comment if they believe it to also be true or different?
Edit: they said they will change this format soon so that we have pretty much all the data in training, mostly as weekly eras, thus meaning this could change relatively soon