eelbrain.BoostingResult.cross_predict

BoostingResult.cross_predict(x=None, data=None, scale='original', name=None)

Predict responses to x using complementary training data

Parameters:

  • x (NDVar | str | Sequence[NDVar | str]) – Predictors used in the original model fit, or a subset thereof. In order for cross-prediction to be accurate, x needs to match the x used in the original fit exactly in cases and time.

  • data (Dataset) – Dataset with predictors. If ds is specified, x can be omitted.

  • scale (Literal['original', 'normalized']) – Return predictions at the scale of the original data (the y supplied to the boosting() function) or at the normalized scale that is used for model fitting (y and x normalized).

  • name (str) – Name for the output NDVar.

Return type:

NDVar

See also

convolve

Simple prediction of linear model

Notes

This function does not adjust the mean across time of predicted responses; subtract the mean in order to compute explained variance.

Examples

Fit a TRF and reproduce the error using the cross-predict function:

trf = boosting(y, x, 0, 0.5, partitions=5, test=1, partition_results=True)
y_pred = trf.cross_predict(x, scale='normalized')

y_normalized = (y - trf.y_mean) / trf.y_scale
y_residual = y_normalized - y_pred
proportion_explained_l1 = 1 - (y_residual.abs().sum('time') / y_normalized.abs().sum('time'))
proportion_explained_l2 = 1 - ((y_residual ** 2).sum('time') / (y_normalized ** 2).sum('time'))