Predict from a brulee_tab_pfn

Predict from a brulee_tab_pfn

Usage

# S3 method for class 'brulee_tab_pfn'
predict(object, new_data, type = NULL, quantile_levels = (1:9)/10, ...)

Arguments

object

A brulee_tab_pfn object.

new_data

A data frame or matrix of new predictors.

type

A single character. The type of predictions to generate. Valid options are:

• "numeric" for numeric predictions (the mean of the predictive distribution).

• "quantile" for quantiles of the predictive distribution.

• "class" for hard class predictions.

• "prob" for soft class predictions (i.e., class probabilities).

The default (NULL) is "class" for classification and "numeric" for regression.

quantile_levels

A numeric vector of probabilities in (0, 1) for type = "quantile".

...

Not used, but required for extensibility.

Value

A tibble of predictions with one row per row of new_data: .pred_class, .pred_{level} columns, .pred, or .pred_quantile (a hardhat::quantile_pred() column), depending on type.

Details

TabPFN runs the network when predicting: each call feeds the stored training rows and new_data through the model.

R torch currently lacks a memory-efficient attention kernel for Apple GPUs that PyTorch (>= 2.13) has, so with device = "mps" attention is computed in chunks of rows to keep it within the GPU's memory.

Examples

if (FALSE) {
fit <- brulee_tab_pfn(mpg ~ ., data = mtcars)
predict(fit, mtcars[1:3, ])
predict(fit, mtcars[1:3, ], type = "quantile", quantile_levels = c(0.1, 0.9))
}