Is there a way to limit running the validation set? I want to make sure my script works and I can limit training with

steps

but I don't see a way to do the same with the validation set.

Hello! I'm not sure I entirely understand the question. Still, in case it can help, you can define the size of your validation and test set, instead of using

n_windows

. Not sure if this is what you are looking for, so let me know if it helps!

Hey hey - let me try and be more clear. I'd like to do a test run of my script. As it stands, by using

steps

I can limit training to a smaller subset of my training data. Is there way to do some the same for validation? As in, not run through the entire validation set so I can speedily test my script?

Oh I see. I never tried it, but could try setting

val_size

to 0?

I'll try that. Thank you

Hi @Isaac. I dont fully understand what you are trying to control here. What method are you using?

fit

or

cross_validation

? In both cases, you only pass one dataframe with continous time series, and you define the length of the validation set with

val_size

, as Marco said. We currently do not support splitting in the

unique_id

dimension.

My bad, let me try and explain better. I have a dataset with a few million unique ID's. My script is dying right now at

predict

so I'm trying to quickly run through

fit

to get straight to

predict

. Since I use a batch size of 128, I can set

steps

to 2 and only "fit" on the first 256 unique ids. However, the validation part of

fit

runs through the entire set of unique ID's every time. I'd like to bypass it, finish the fitting and go straight into predicting. Is there a way to do that? Does that make sense?

what model are you using?

val_size=0

plus

num_sanity_val_steps=0

should work

I'm testing a few models, AutoLSTM, AutoNHITS, AutoTFT