In ​*NeuralForecast*​, your approach is correct but there's an important distinction to understand about how

Auto

models work with cross-validation (1) . When you use

Auto

models like

AutoNHITS

or

AutoLSTM

, the hyperparameter optimization happens during the initial

.fit()

method . The

Auto

models automatically perform hyperparameter tuning, training multiple models with different hyperparameters, producing forecasts on the validation set, and evaluating them . Only the best model is stored and used during inference . However, when you subsequently call

cross_validation

, you're not re-optimizing hyperparameters . The cross-validation method uses the already-optimized model with the best hyperparameters found during the initial training phase . The cross-validation process allows you to simulate multiple historic forecasts by defining a sliding window across the historical data and predicting the period following it . This gives you a better estimation of your model's predictive abilities across a wider range of temporal instances . For model comparison across your 7 time series, you can use the evaluation approach shown in the documentation :

from utilsforecast.evaluation import evaluate
from utilsforecast.losses import mse, mae, rmse

evaluation_df = evaluate(cv_df.drop(columns='cutoff') , metrics=[mse, mae, rmse])
evaluation_df['best_model'] = evaluation_df.drop(columns=['metric', 'unique_id']) .idxmin(axis=1)

This will help you determine which model performs best for each individual

unique_id

.

To learn more, see the sources I considered: [[1] End to End Walkthrough](https://nixtlaverse.nixtla.io/neuralforecast/docs/tutorials/getting_started_complete.html)

Thanks for the feedback, glad I was helpful.