pmdarima (Auto-ARIMA)

2.1.1 · active · verified Thu Apr 09

pmdarima is a Python library that provides an equivalent to R's `auto.arima` function, automating the process of selecting optimal ARIMA (AutoRegressive Integrated Moving Average) models for time series forecasting. It builds on `statsmodels` but offers a scikit-learn-like API, simplifying complex time series analysis. The library is currently at version 2.1.1 and receives regular updates for Python version compatibility and dependency support.

Warnings

breaking Version 2.1.0 removed support for Python 3.7, 3.8, and 3.9. Projects using these older Python versions must either remain on pmdarima < 2.1.0 or upgrade their Python environment.
Fix: Upgrade Python to 3.10 or newer. As of v2.1.1, Python 3.10, 3.11, 3.12, 3.13, and 3.14 are supported.
breaking Version 2.1.0 introduced support for Numpy 2.x and simultaneously removed support for Numpy 1.x. Older projects might encounter build or runtime errors if Numpy is not updated.
Fix: Ensure Numpy is upgraded to version 2.0.0 or newer.
breaking Version 2.1.0 increased the minimum required versions for SciPy to >=1.13.0 and Statsmodels to >=0.14.5. Using older versions of these dependencies will likely result in installation issues or runtime errors.
Fix: Update SciPy to 1.13.0 or higher and Statsmodels to 0.14.5 or higher.
gotcha The `m` parameter (seasonal period) in `auto_arima` is not automatically detected and must be specified by the user. An incorrect `m` value can lead to suboptimal or erroneous seasonal models.
Fix: Carefully determine the seasonal period (`m`) based on the data's characteristics (e.g., 12 for monthly data, 4 for quarterly data, 7 for daily data with weekly seasonality). Seasonal decomposition (`pmdarima.arima.model.tsdisplay`) or domain knowledge can help.
deprecated The `exogenous` and `sarimax_kwargs` arguments to `ARIMA` and `auto_arima` were deprecated in earlier 1.x versions and will now raise a `TypeError` if used in 2.0.0 and later.
Fix: Replace `exogenous` with the `X` parameter and `sarimax_kwargs` with direct keyword arguments passed to the underlying `SARIMAX` model via `**kwargs`.

Install

pip install pmdarima Install pmdarima

Imports

auto_arima
```
from pmdarima import auto_arima
```
ARIMA
```
from pmdarima.arima import ARIMA
```
ARIMA is located in the `pmdarima.arima` submodule, not directly under `pmdarima`.

Quickstart

This quickstart demonstrates how to use `pmdarima.auto_arima` to automatically select and fit an ARIMA model to a time series and generate future predictions. The example uses a simple synthetic dataset, fits the model, and then forecasts 10 future periods. Parameters like `start_p`, `start_q`, `max_p`, `max_q`, and `seasonal` control the search space for the optimal ARIMA model.

import pmdarima as pm
import numpy as np
import matplotlib.pyplot as plt

# Generate some sample time series data
y = np.random.rand(100) * 10 + np.arange(100) # Simple trend + noise

# Fit a stepwise auto_arima model
model = pm.auto_arima(y, 
                      start_p=1, start_q=1,
                      test='adf',       # use adftest to find optimal 'd'
                      max_p=3, max_q=3, # maximum p and q
                      m=1,              # frequency of series
                      d=None,           # let model determine 'd'
                      seasonal=False,   # No seasonality
                      start_P=0, 
                      D=0,
                      trace=False,      # Suppress verbose output
                      error_action='ignore',  
                      suppress_warnings=True, 
                      stepwise=True)

# Make predictions
forecast, conf_int = model.predict(n_periods=10, return_conf_int=True)

print("Forecast:", forecast)
print("Confidence Interval:", conf_int)

# Optional: plot results
# plt.plot(y, label='Actual')
# plt.plot(np.arange(len(y), len(y) + len(forecast)), forecast, label='Forecast')
# plt.fill_between(np.arange(len(y), len(y) + len(forecast)),
#                  conf_int[:, 0], conf_int[:, 1], alpha=0.1)
# plt.legend()
# plt.show()

view raw JSON →