TabPFN Common Utilities

A comprehensive utility package for TabPFN - the foundation model for tabular data.

Features

🔒 Privacy-First Telemetry System

• Anonymous & Aggregated Data Collection: Implements safe, GDPR-compliant telemetry that respects user privacy
• Configurable Analytics: Optional telemetry that can be disabled via environment variables
• Usage Pattern Insights: Tracks TabPFN usage patterns to improve the model and user experience
• Zero Personal Data: No personal information or sensitive data is collected or transmitted

💰 Cost Estimation

• Resource Planning: Accurate estimation of computational costs and duration for TabPFN predictions
• Cloud Pricing: Essential for resource planning in cloud-based TabPFN services
• Task-Specific Calculations: Different cost models for classification vs regression tasks

📊 Data Processing Utilities

• Regression Results: Comprehensive handling of prediction outputs with mean, median, mode, and quantiles
• Data Serialization: Convert between pandas DataFrames, NumPy arrays, and CSV formats
• Dataset Management: Load and preprocess standard ML datasets with proper train/test splits
• Preprocessing Configuration: Extensive options for data transformation strategies

Installation

pip install tabpfn-common-utils

Or with uv:

uv add tabpfn-common-utils

Quick Start

Telemetry (Privacy-Compliant)

from tabpfn_common_utils.telemetry import ProductTelemetry

# Initialize telemetry service (anonymous, GDPR-compliant)
telemetry = ProductTelemetry()

# Track usage events (no personal data collected)
telemetry.capture(...)

# Telemetry can be disabled by setting environment variable
export TABPFN_DISABLE_TELEMETRY=1

Regression Results

from tabpfn_common_utils.regression_pred_result import RegressionPredictResult

# Handle regression prediction results
result = RegressionPredictResult({
    "mean": [1.2, 2.3, 3.4],
    "median": [1.1, 2.2, 3.3],
    "mode": [1.0, 2.0, 3.0],
    "quantile_0.25": [0.9, 1.9, 2.9],
    "quantile_0.75": [1.5, 2.5, 3.5]
})

# Convert to basic representation for serialization
basic_repr = RegressionPredictResult.to_basic_representation(result)

Data Utilities

from tabpfn_common_utils.utils import get_example_dataset, serialize_to_csv_formatted_bytes
import pandas as pd

# Load example dataset
X_train, X_test, y_train, y_test = get_example_dataset("iris")

# Serialize data to CSV bytes
csv_bytes = serialize_to_csv_formatted_bytes(X_train)

Privacy & Compliance

This package implements privacy-first telemetry that:

• ✅ GDPR Compliant: No personal data collection
• ✅ Anonymous Only: No user identification or tracking
• ✅ Aggregated Data: Only statistical insights are collected
• ✅ User Control: Can be completely disabled
• ✅ Transparent: Open source code for full transparency

Telemetry data helps improve TabPFN but never compromises user privacy.

Development

Setup

# Install dependencies
uv sync

# Activate virtual environment
source .venv/bin/activate

# Run tests
uv run pytest

# Type checking
uv run pyright

# Code formatting
uv run ruff check --fix

Adding Dependencies

# Add runtime dependency
uv add <package_name>

# Add development dependency
uv add --group dev <package_name>

📝 License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contributing

Contributions are welcome! Please ensure all code passes type checking and formatting requirements.

Links

• TabPFN Main Repository
• Documentation
• Issues