Skip to main content

​
TabPFN-TS

TabPFN-TS is a specialized variant of TabPFN designed specifically for time series forecasting and temporal data analysis.

​
Time Series Specialization

  • Temporal Patterns: Optimized for capturing time-dependent relationships
  • Seasonality Detection: Automatic identification of seasonal patterns
  • Trend Analysis: Robust handling of various trend types
  • Missing Data: Intelligent imputation for incomplete time series

​
Key Advantages

  • No Training Required: Instant forecasts without model training
  • Multiple Horizons: Predict multiple time steps ahead simultaneously
  • Uncertainty Quantification: Built-in confidence intervals
  • Multiple Series: Handle multiple time series in a single call

​
Supported Patterns

Seasonal Patterns

Daily, weekly, monthly, and yearly seasonality

Trend Analysis

Linear, exponential, and polynomial trends

Anomaly Detection

Automatic outlier identification

Multi-Variate

Handle multiple correlated time series

​
Usage Example

from tabpfn import TabPFNTimeSeries
import pandas as pd

# Initialize time series model
ts_model = TabPFNTimeSeries()

# Your time series data
data = pd.read_csv("timeseries_data.csv")
data['date'] = pd.to_datetime(data['date'])
data = data.set_index('date')

# Forecast next 30 days
forecast = ts_model.forecast(
    data=data,
    horizon=30,
    confidence_level=0.95
)

print(f"Forecast: {forecast.predictions}")
print(f"Confidence Intervals: {forecast.confidence_intervals}")

​
Performance Metrics

TabPFN-TS excels in time series forecasting:
  • MAPE: 15% average improvement over traditional methods
  • RMSE: 20% reduction in root mean square error
  • Speed: 100x faster than training-based approaches
  • Memory: Minimal memory footprint for large datasets
⌘I