Hello, I'm trying to implement an auto NBEATS model and crossvalidate it. When running the crossvalidation it appears to work but then suddenly stops and displays this error message:

*TypeError*: TimeSeriesDataModule.__init__() got an unexpected keyword argument 'config'

. When using the ray backend another error happens as well. I also tried fitting the model outside of crossvallidation and got this:

*AttributeError*: 'DataFrame' object has no attribute 'temporal_cols'

Not really sure what is going on, I'm running this on anaconda3 in python3.10.4. Any help is appreciated. This is my code:

windows = (validation_size - horizon + 1) // horizon
cv_results = nf.cross_validation(df, n_windows=windows, step_size=horizon, config = None)


horizon = 12    
nf = NeuralForecast(
    models=[
        AutoNBEATS(h=horizon, num_samples=5, gpus=1, loss=MQLoss(), backend='optuna', alias="NBEATS ray "),
    ],  
    freq='h'
        )  
windows = (validation_size - horizon + 1) // horizon
cv_results = nf.cross_validation(df, n_windows=windows, step_size=horizon, config = None)

Why are you passing

config=None

to cross_validation?

To get the standard config, is that wrong?

That's an argument of the model, so it should go inside

AutoNBEATS

not in the

cross_validation

call

Oh my god, how did not see that. So sorry. I'll run it again and see if anything else pops up

no worries. let us know if you run into any problems

It ran fine thanks, any ideas why this happens:

*AttributeError*: 'DataFrame' object has no attribute 'temporal_cols'

What's the code that you're running when you get that error?

It seems I forgot to paste it, this one here:

model = AutoNBEATS(h=12, num_samples=5, gpus=1, loss=MQLoss(), backend='optuna', alias="NBEATS ray ")
model.fit(dataset=df)

you have to put that model in a NeuralForecast object and call fit on that, as you do with the cross_validation above

I see, makes sense. I think there might be a mistake in the docs then. In the AutoNBEATS example it uses it like I have.

That example builds the dataset at the top:

# Split train/test and declare time series dataset
Y_train_df = Y_df[Y_df.ds<='1959-12-31'] # 132 train
Y_test_df = Y_df[Y_df.ds>'1959-12-31']   # 12 test
dataset, *_ = TimeSeriesDataset.from_df(Y_train_df)

I missed that part, my mistake. Thanks for the tip and thank you so much for the help.