Sampling

If your dataset does not fit in your RAM, you may want to subsample the set on which splitting will be performed.

Partitions can be selected from the "Target" tab.

Sampling

Time series resampling

Data will be resampled with a frequency (time step) of {{ prettyTimeSteps(mlTaskDesign.timestepParams.numberOfTimeunits, mlTaskDesign.timestepParams.timeunit) }}

Numerical interpolation

Numerical extrapolation

Extrapolation start date

First available Custom

Numerical features are only inferred when no known data is before this date. This date is ignored otherwise.

Extrapolation end date

Last available Custom

Numerical features are only inferred when no known data is after this date. This date is ignored otherwise.

Non-numerical imputation

Duplicate timestamp handling

{{ prettyTimeSteps(uiState.numberOfHorizonsInTest * uiState.predictionLength * uiState.timestepParams.numberOfTimeunits, uiState.timestepParams.timeunit) }}

K-fold cross-test

Gives error margins on metrics, but greatly increases training time

Number of folds

Number of folds to divide the dataset into

Fold offset

Add an offset of one forecast horizon between the cross-test evaluation (resp. hyperparameter search validation) folds

Equal duration train set folds

Create rolling train set folds of equal duration.

Sampling

Sampling

Time series resampling

Metadata