Skip to content

TabPFN: prior data fitted networks

Source: R/tabular_pfn.R
tabular_pfn.Rd

This function uses a pre-trained deep learning network that emulates Bayesian inference. The model was trained on a large number of simulated data sets and an attention mechanism is use to make relevant predictions for specific (i.e., real) data sets.

There are different ways to fit this model, and the method of estimation is chosen by setting the model engine. The engine-specific pages for this model are listed below.

¹ The default engine. ² Requires a parsnip extension package for classification and regression.

More information on how parsnip is used for modeling is at https://www.tidymodels.org/.

Usage

tabular_pfn(
  mode = "unknown",
  engine = "tabpfn",
  num_estimators = NULL,
  softmax_temperature = NULL,
  balance_probabilities = NULL,
  average_before_softmax = NULL
)

Arguments

mode

A single character value for the type of model. The possible values for this model are "classification" and "regression".

engine

A single character string specifying what computational engine to use for fitting. Possible engines are listed below. The default for this model is "tabpfn".

num_estimators

An integer for the ensemble size. Default is 8L.

softmax_temperature

An adjustment factor that is a divisor in the exponents of the softmax function (see tabpfn::tab_pfn()). Defaults to 0.9.

balance_probabilities

A logical to adjust the prior probabilities in cases where there is a class imbalance. Default is FALSE. Classification only.

average_before_softmax

A logical. For cases where num_estimators > 1, should the average be done before using the softmax function or after? Default is FALSE.

Details

This function fits classification and regression models.

References

https://github.com/PriorLabs/TabPFN

Hollmann, Noah, Samuel Müller, Lennart Purucker, Arjun Krishnakumar, Max Körfer, Shi Bin Hoo, Robin Tibor Schirrmeister, and Frank Hutter. "Accurate predictions on small data with a tabular foundation model." Nature 637, no. 8045 (2025): 319-326.

Hollmann, Noah, Samuel Müller, Katharina Eggensperger, and Frank Hutter. "Tabpfn: A transformer that solves small tabular classification problems in a second." arXiv preprint arXiv:2207.01848 (2022).

Müller, Samuel, Noah Hollmann, Sebastian Pineda Arango, Josif Grabocka, and Frank Hutter. "Transformers can do Bayesian inference." arXiv preprint arXiv:2112.10510 (2021).

See also

fit(), set_engine(), update(), tabpfn engine details tabpfn::tab_pfn()

Examples

show_engines("tabular_pfn")
#> # A tibble: 0 × 2
#> # ℹ 2 variables: engine <chr>, mode <chr>

tabular_pfn()
#> ! parsnip could not locate an implementation for `tabular_pfn` model
#>   specifications.
#> ℹ The parsnip extension package tabby implements support for this
#>   specification.
#> ℹ Please install (if needed) and load to continue.
#> tabular pfn Model Specification (unknown mode)
#> 
#> Computational engine: tabpfn 
#>