Cross Validation¶

The number of data for traing should be between 70-80% of the total historical data. If we have less data, perhaps you should use more data (up to 90%) for training.

purpose¶

• evaluate how well a model will perform on unseen data, helping to detect and prevent overfitting

• used to select the model parameters with the best score in hyperparameter tuning

• ts-cross-val: efficiently use all training data to get the best performance evaluation

sliding window¶

The total training and test data points are the same but move to the future.

expanding window¶

The training window expands while the test window slides.