Hi channel, I'm currently facing two errors (sometimes I run the cell i get one, another times I get the other): The first picture is my instantiation of the MLForecast plus the model. The second picture is my most common error. The third picture is another that sometimes comes instead of the second one. NOTE: these errors only happen when I give the parameter max_horizon = any int(>2), if i dont give it, or if i give only 2 it runs without errors. Did anyone faced the same issue? PS: I tried running the one-model-per-step tutorial, and gives exactly the same error.

What do you get if you run

train[id_col].value_counts().min()

?

1

And

train[id_col].value_counts().max()

?

I think the problem may be the max_horizon. It needs to train a model to predict 6 steps ahead but you have series with only one sample. Do you get the same error if you remove it?

no.. only when i apply max horizon

And what's the max lag that you're using? Since you're setting

dropna=True

it's possible that it's dropping several samples in each serie

LAGS = [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111]

Yeah so you'd end up with 2 samples in each serie and you need at least 6. Can you try setting

dropna=False

or reducing the maximum lag?

LAGS = [ 1, 2, 3, 4, 5, 6] dropna=False

Hmm. Did you define a Dataset object in your notebook?

Not sure what you mean? I always feed it pandas dfs

I mean something like:

Dataset = something

or

class Dataset:

no... simply did train = pd.read_csv(...)

Can you restart your kernel? It seems something corrupted LightGBM's Dataset

ok... but just fyi i'm working on 4 different notebooks in 4 different machines. Colab, Sagemaker, Vm, Local machine.

We'll add a better error message for that case

Thank you... and thank you for the help 🙌

Yes, assuming all of your series start and end at the same timestamp, otherwise you should check the sizes

yes yes.. assuming the dates are equal for all unique_ids