StatsForecast

2.0.3 · active · verified Fri Apr 10

StatsForecast is a Python library providing a lightning-fast suite of statistical and econometric models for time series forecasting. It offers highly optimized implementations of models like ARIMA, ETS, CES, and Theta, designed for speed and scalability to forecast millions of series efficiently. The library is currently at version 2.0.3 and maintains an active release cadence with frequent updates and performance enhancements.

Warnings

breaking The default values for `allowmean` and `allowdrift` in `AutoARIMA` changed from `False` to `True` in `v2.0.0`. This can alter the behavior and results of your `AutoARIMA` forecasts if you were relying on the previous defaults.
Fix: Explicitly set `allowmean=False` and/or `allowdrift=False` in `AutoARIMA` constructor if you need the old behavior, or review your models to account for the new defaults.
breaking The `df` argument has been removed from the `StatsForecast` class constructor as of `v2.0.3`. Instead, the DataFrame should be passed directly to the `fit`, `forecast`, or `cross_validation` methods.
Fix: Remove the `df` argument from the `StatsForecast()` constructor. Pass your DataFrame directly to methods like `sf.fit(df)` or `sf.forecast(df=df, h=...)`.
deprecated Pandas frequency alias 'H' (for hourly data) is deprecated and will be removed in a future version. Use 'h' instead.
Fix: Replace `freq='H'` with `freq='h'` when defining the frequency in `StatsForecast` or other pandas-related operations.
breaking Version 2.0.0 included a general breaking change to 'remove deprecated behavior' as part of refactoring. While specific impacts vary, it signals that code using previously deprecated features will likely break.
Fix: Consult the official release notes for v2.0.0 and update code using features that were marked as deprecated in earlier versions.

Install

pip install statsforecast PyPI

Imports

StatsForecast
```
from statsforecast import StatsForecast
```

AutoARIMA

from statsforecast.models import AutoARIMA

AirPassengersDF
```
from statsforecast.utils import AirPassengersDF
```
AirPassengersDF is a utility dataset moved to `statsforecast.utils`.

Quickstart

This quickstart demonstrates how to use StatsForecast to fit an AutoARIMA model to the classic AirPassengers dataset and generate future predictions with confidence intervals. The input DataFrame must be in a 'long' format with 'unique_id', 'ds' (datestamp), and 'y' (target) columns.

import pandas as pd
from statsforecast import StatsForecast
from statsforecast.models import AutoARIMA
from statsforecast.utils import AirPassengersDF

# Load example data (AirPassengers dataset)
df = AirPassengersDF

# Instantiate StatsForecast with models and frequency
# For monthly data, 'M' or 'ME' (MonthEnd) is common
sf = StatsForecast(
    models=[AutoARIMA(season_length=12)],
    freq='M',
    n_jobs=-1 # Use all available cores for parallel processing
)

# Fit the models
sf.fit(df)

# Make predictions for the next 12 steps (horizon=12)
# and calculate 95% prediction intervals
forecast_df = sf.predict(h=12, level=[95])

print(forecast_df.head())

view raw JSON →