I don't understand what dataset format you use to train the model

[ammarsaf]

I used MLForecast tutorial from here, but when I started training it raise the error of

ValueError: Found array with 0 sample(s) (shape=(0, 2)) while a minimum of 1 is required by LinearRegression.

(as I removed the LightGBM as I thought it was the problem, but the issue is the same, array is empty)

I remove the RollingMean part, make the model much simpler, but still did not solved the issue.

models = [
    # lgb.LGBMRegressor(random_state=32, verbosity=1), 
    LinearRegression()
    
]

model_mlfc = MLForecast(
    models=models, 
    freq="D", 
    lags=[7], 
    date_features=["dayofweek"], 
    target_transforms=[Differences([1])]
) 

Here is my dataset from kaggle

Do we need to make the dataset to be time indexed first by groupby date with mean/sum, or make it raw?

How exactly the unique id should be?

Hoe exactly you need the dataset to be processed before it can be passed to the model? I hope you can state it in the documentation.

[ammarsaf]

Okay I solved it. The unique ID should be a single number since it coming form a single series.