TabPFN Classification Learner

Foundation model for tabular data. Uses reticulate to interface with the tabpfn Python package.

Installation

This learner relies on reticulate to handle Python dependencies. It is not necessary to install any Python package manually in advance or specify a Python environment via reticulate::use_python(), reticulate::use_virtualenv(), reticulate::use_condaenv(), or reticulate::use_miniconda(). By calling $train() or $predict(), the required Python packages (tabpfn, torch, etc.) will be installed automatically, if not already. Reticulate will then configure and initialize an ephemeral environment satisfying those requirements, unless an existing environment (e.g., "r-reticulate") in reticulate's Order of Discovery contains all the necessary packages.

You may also manually install tabpfn into a Python environment following the official installation guide and specify the environment via reticulate::use_*() before calling $train() or $predict(). Note that the GPU version of PyTorch cannot be loaded by reticulate::use_condaenv() from a conda environment. To use a conda environment, please install the CPU version of PyTorch.

License acceptance

Since tabpfn version 8, the default model version (v3) requires a one-time acceptance of the Prior Labs license before the model weights can be downloaded. On the first call to $train() or $predict(), tabpfn opens a browser window to log in and accept the license, and caches the resulting token for subsequent calls. In non-interactive environments (e.g., on a server or in continuous integration), this browser flow fails. Instead, log in at https://ux.priorlabs.ai, accept the license on the licenses tab, copy your API key from https://ux.priorlabs.ai/account, and set it as the environment variable TABPFN_TOKEN before training.

Saving a Learner

In order to save a LearnerClassifTabPFN for later usage, it is necessary to call the $marshal() method on the Learner before writing it to disk, as the object will otherwise not be saved correctly. After loading a marshaled LearnerClassifTabPFN into R again, you then need to call $unmarshal() to transform it into a useable state.

Initial parameter values

• n_jobs is initialized to 1 to avoid threading conflicts with future.

Custom mlr3 parameters

• categorical_feature_indices uses R indexing instead of zero-based Python indexing.

• device must be a string. If set to "auto", the behavior is the same as original. Otherwise, the string is passed as argument to torch.device() to create a device.

• inference_precision must be "auto", "autocast", or a torch.dtype string, e.g., "torch.float32", "torch.float64", etc. Non-float dtypes are not supported.

• inference_config is currently not supported.

• tuning_config must be passed as a named list, which is converted to a Python dictionary. It enables post-hoc tuning towards eval_metric and accepts the keys calibrate_temperature, tune_decision_thresholds, tuning_holdout_frac, and tuning_n_folds.

• eval_metric only affects predictions when tuning_config is also set, in which case the softmax temperature and decision thresholds are tuned towards the chosen metric during training.

• n_jobs is deprecated upstream in favor of n_preprocessing_jobs and is only kept for backward compatibility.

• random_state accepts either an integer or the special value "None" which corresponds to None in Python. Following the original Python implementation, the default random_state is 0.

Dictionary

This Learner can be instantiated via lrn():

lrn("classif.tabpfn")

Meta Information

• Task type: “classif”

• Predict Types: “response”, “prob”

• Feature Types: “logical”, “integer”, “numeric”

• Required Packages: mlr3, reticulate

Parameters

Id	Type	Default	Levels	Range
n_estimators	integer	8		\([1, \infty)\)
auto_scale_n_estimators	logical	TRUE	TRUE, FALSE	-
categorical_features_indices	untyped	-		-
softmax_temperature	numeric	0.9		\([0, \infty)\)
balance_probabilities	logical	FALSE	TRUE, FALSE	-
average_before_softmax	logical	FALSE	TRUE, FALSE	-
model_path	untyped	"auto"		-
device	untyped	"auto"		-
ignore_pretraining_limits	logical	FALSE	TRUE, FALSE	-
inference_precision	character	auto	auto, autocast, torch.float32, torch.float, torch.float64, torch.double, torch.float16, torch.half, torch.bfloat16	-
fit_mode	character	fit_preprocessors	low_memory, fit_preprocessors, fit_with_cache, batched	-
memory_saving_mode	untyped	"auto"		-
keep_cache_on_device	logical	TRUE	TRUE, FALSE	-
random_state	integer	0		\((-\infty, \infty)\)
n_jobs	integer	-		\([1, \infty)\)
n_preprocessing_jobs	integer	1		\([1, \infty)\)
differentiable_input	logical	FALSE	TRUE, FALSE	-
eval_metric	character	-	accuracy, balanced_accuracy, roc_auc, f1, log_loss	-
tuning_config	untyped	-		-
show_progress_bar	logical	FALSE	TRUE, FALSE	-

References

Hollmann, Noah, Müller, Samuel, Purucker, Lennart, Krishnakumar, Arjun, Körfer, Max, Hoo, Bin S, Schirrmeister, Tibor R, Hutter, Frank (2025). “Accurate predictions on small data with a tabular foundation model.” Nature. doi:10.1038/s41586-024-08328-6 . https://www.nature.com/articles/s41586-024-08328-6.

Hollmann, Noah, Müller, Samuel, Eggensperger, Katharina, Hutter, Frank (2023). “TabPFN: A transformer that solves small tabular classification problems in a second.” In International Conference on Learning Representations 2023.

See also

• Dictionary of Learners: mlr3::mlr_learners.

• as.data.table(mlr_learners) for a table of available Learners in the running session (depending on the loaded packages).

• Chapter in the mlr3book: https://mlr3book.mlr-org.com/chapters/chapter2/data_and_basic_modeling.html#sec-learners

• mlr3learners for a selection of recommended learners.

• mlr3cluster for unsupervised clustering learners.

• mlr3pipelines to combine learners with pre- and postprocessing steps.

• mlr3tuning for tuning of hyperparameters, mlr3tuningspaces for established default tuning spaces.

Author

b-zhou

Super classes

mlr3::Learner -> mlr3::LearnerClassif -> LearnerClassifTabPFN

Active bindings

marshaled

(logical(1))
Whether the learner has been marshaled.

Methods

Public methods

• LearnerClassifTabPFN$new()

• LearnerClassifTabPFN$marshal()

• LearnerClassifTabPFN$unmarshal()

• LearnerClassifTabPFN$clone()

Inherited methods

• mlr3::Learner$base_learner()
• mlr3::Learner$configure()
• mlr3::Learner$encapsulate()
• mlr3::Learner$format()
• mlr3::Learner$help()
• mlr3::Learner$predict()
• mlr3::Learner$predict_newdata()
• mlr3::Learner$print()
• mlr3::Learner$reset()
• mlr3::Learner$selected_features()
• mlr3::Learner$train()
• mlr3::LearnerClassif$predict_newdata_fast()

---

LearnerClassifTabPFN$new()

Creates a new instance of this R6 class.

Usage

LearnerClassifTabPFN$new()

---

LearnerClassifTabPFN$marshal()

Marshal the learner's model.

Usage

LearnerClassifTabPFN$marshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::marshal_model().

---

LearnerClassifTabPFN$unmarshal()

Unmarshal the learner's model.

Usage

LearnerClassifTabPFN$unmarshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::unmarshal_model().

---

LearnerClassifTabPFN$clone()

The objects of this class are cloneable with this method.

Usage

LearnerClassifTabPFN$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.