Skip to main content

​
softmax_temperature

Controls prediction sharpness (classification only):
  • Lower values (e.g., 0.7): sharper, more confident predictions — useful when accuracy is already high
  • Higher values (e.g., 1.2): softer, more calibrated predictions — useful when probability calibration matters
model = TabPFNClassifier(softmax_temperature=0.8)
If you use tuning_config={"calibrate_temperature": True}, the temperature is tuned automatically and overrides this value.

​
Metric Tuning

For metrics that are sensitive to decision thresholds (F1, balanced accuracy, precision, recall), use the built-in metric tuning: 
model = TabPFNClassifier(
    eval_metric="f1",
    tuning_config={
        "calibrate_temperature": True,
        "tune_decision_thresholds": True,
    },
)

​
Handling Imbalanced Data

  • Set balance_probabilities=True as a quick heuristic for imbalanced datasets when your evaluation metric weights each class equally regardless of its frequency (e.g. balanced accuracy, balanced log loss).
  • For more control, use eval_metric="balanced_accuracy" with threshold tuning
balance_probabilities does not always help. In some cases it can balance predictions at the cost of overall predictive power. Test both settings.

​
n_estimators and auto_scale_n_estimators

n_estimators controls the number of ensemble members (default 8). Each estimator uses a different preprocessing configuration, contributing to ensemble diversity and robustness. By default (auto_scale_n_estimators=True), TabPFN automatically increases n_estimators on wide datasets to ensure full feature coverage — the actual count is capped at 32. To use exactly the number you specify without any auto-scaling, set auto_scale_n_estimators=False: 
from tabpfn import TabPFNClassifier

# Disable auto-scaling — use exactly 8 estimators regardless of feature count
model = TabPFNClassifier(n_estimators=8, auto_scale_n_estimators=False)
Disabling auto-scaling is useful when you need a fixed, predictable computational budget or when comparing models with a controlled number of estimators.