many_class_classifier ¶

This module provides a classifier that overcomes TabPFN's limitation on the number of classes (typically 10) by using a meta-classifier approach. It works by breaking down multi-class problems into multiple sub-problems, each within TabPFN's class limit.

Key features: - Handles any number of classes beyond TabPFN's native limits - Uses an efficient codebook approach to minimize the number of base models - Compatible with both TabPFN and TabPFN-client backends - Maintains high accuracy through redundant coding - Follows scikit-learn's Estimator interface

Example usage

from tabpfn import TabPFNClassifier  # or from tabpfn_client
from tabpfn_extensions.many_class import ManyClassClassifier

# Create a base TabPFN classifier
base_clf = TabPFNClassifier()

# Wrap it with ManyClassClassifier to handle more classes
many_class_clf = ManyClassClassifier(
    estimator=base_clf,
    alphabet_size=10  # Use TabPFN's maximum class limit
)

# Use like any scikit-learn classifier, even with more than 10 classes
many_class_clf.fit(X_train, y_train)
y_pred = many_class_clf.predict(X_test)

ManyClassClassifier ¶

Bases: BaseEstimator, ClassifierMixin

Output-Code multiclass strategy to extend TabPFN beyond its class limit.

This class enables TabPFN to handle classification problems with any number of classes by using a meta-classifier approach. It creates an efficient coding system that maps the original classes to multiple sub-problems, each within TabPFN's class limit.

Parameters:

Name	Description	Default
`estimator`	A classifier implementing fit() and predict_proba() methods. Typically a TabPFNClassifier instance. The base classifier should be able to handle up to `alphabet_size` classes.	required
`alphabet_size`	Maximum number of classes the base estimator can handle. If None, it will try to get this from estimator.max_num_classes_	`None`
`n_estimators`	Number of base estimators to train. If None, an optimal number is calculated based on the number of classes and alphabet_size.	`None`
`n_estimators_redundancy`	Redundancy factor for the auto-calculated number of estimators. Higher values increase reliability but also computational cost.	`4`
`random_state`	Controls randomization used to initialize the codebook. Pass an int for reproducible results.	`None`

Attributes:

Name	Type	Description
`classes_`		Array containing unique target labels.
`code_book_`		N-ary array containing the coding scheme that maps original classes to base classifier problems.
`no_mapping_needed_`		True if the number of classes is within the alphabet_size limit, allowing direct use of the base estimator without any mapping.
`classes_index_`		Maps class labels to their indices in classes_.
`X_train`		Training data stored for reuse during prediction.
`Y_train`		Encoded training labels for each base estimator.

Examples:

>>> from sklearn.datasets import load_iris
>>> from tabpfn import TabPFNClassifier
>>> from tabpfn_extensions.many_class import ManyClassClassifier
>>> from sklearn.model_selection import train_test_split
>>> X, y = load_iris(return_X_y=True)
>>> X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
>>> base_clf = TabPFNClassifier()
>>> many_clf = ManyClassClassifier(base_clf, alphabet_size=base_clf.max_num_classes_)
>>> many_clf.fit(X_train, y_train)
>>> y_pred = many_clf.predict(X_test)

fit ¶

fit(
    X: ndarray, y: ndarray, **fit_params
) -> ManyClassClassifier

Fit underlying estimators.

Parameters:

Name	Type	Description	Default
`X`	`ndarray`	Data matrix of shape (n_samples, n_features).	required
`y`	`ndarray`	Multi-class target labels of shape (n_samples,).	required
`**fit_params`		Parameters passed to the `estimator.fit` method of each sub-estimator.	`{}`

Returns:

Name	Type	Description
`self`	`ManyClassClassifier`	Returns a fitted instance of self.

get_alphabet_size ¶

get_alphabet_size() -> int

Get the alphabet size to use for the codebook.

Returns:

Name	Type	Description
`int`	`int`	The alphabet size to use

get_n_estimators ¶

get_n_estimators(n_classes: int) -> int

Calculate the number of estimators to use based on the number of classes.

Parameters:

Name	Type	Description	Default
`n_classes`	`int`	The number of classes in the classification problem	required

Returns:

Name	Type	Description
`int`	`int`	The number of estimators to use

predict ¶

predict(X: ndarray) -> ndarray

Predict multi-class targets using underlying estimators.

Parameters:

Name	Type	Description	Default
`X`	`ndarray`	Data matrix of shape (n_samples, n_features).	required

Returns:

Type	Description
`ndarray`	np.ndarray: Predicted multi-class targets of shape (n_samples,).

predict_proba ¶

predict_proba(X: ndarray) -> ndarray

Predict probabilities using the underlying estimators.

Parameters:

Name	Type	Description	Default
`X`	`ndarray`	Data matrix of shape (n_samples, n_features).	required

Returns:

Type	Description
`ndarray`	np.ndarray: The probability of the samples for each class in the model, of shape (n_samples, n_classes), where classes are ordered as they are in `self.classes_`.

set_categorical_features ¶

set_categorical_features(
    categorical_features: list[int],
) -> None

Set categorical features for the base estimator if it supports it.

Parameters:

Name	Type	Description	Default
`categorical_features`	`list[int]`	List of categorical feature indices.	required