Skip to content

Getting Started

In this tutorial, we will install Yohou-Nixtla, fit a forecaster to the classic Air Passengers dataset, and generate a 12-month forecast.

Install Yohou-Nixtla

pip install yohou-nixtla

uv add yohou-nixtla

Verify the installation:

import yohou_nixtla
print(yohou_nixtla.__version__)

The output should look something like:

0.1.0

Load the data

We will use the Air Passengers dataset, a monthly time series of international airline passengers from 1949 to 1960. Yohou ships it as a built-in dataset:

from yohou.datasets import load_air_passengers

y = load_air_passengers()
print(y.head(3))

shape: (3, 2)
┌─────────────────────┬────────────┐
│ time                ┆ Passengers │
│ ---                 ┆ ---        │
│ datetime[μs]        ┆ i64        │
╞═════════════════════╪════════════╡
│ 1949-01-01 00:00:00 ┆ 112        │
│ 1949-02-01 00:00:00 ┆ 118        │
│ 1949-03-01 00:00:00 ┆ 132        │
└─────────────────────┴────────────┘

Notice the two columns: time holds timestamps, and Passengers holds the values. This is Yohou's standard wide-format layout.

Now split the data, keeping the last 12 months as a holdout:

y_train = y.head(len(y) - 12)
y_test = y.tail(12)

Fit a forecaster

Create an AutoARIMAForecaster and fit it on the training data. season_length=12 tells the model to look for annual seasonality in monthly data:

from yohou_nixtla import AutoARIMAForecaster

forecaster = AutoARIMAForecaster(season_length=12)
forecaster.fit(y_train, forecasting_horizon=12)

Generate predictions

Predict the next 12 months:

y_pred = forecaster.predict(forecasting_horizon=12)
print(y_pred.head(3))

shape: (3, 3)
┌─────────────────────┬─────────────────────┬────────────┐
│ observed_time       ┆ time                ┆ Passengers │
│ ---                 ┆ ---                 ┆ ---        │
│ datetime[μs]        ┆ datetime[μs]        ┆ i64        │
╞═════════════════════╪═════════════════════╪════════════╡
│ 1959-12-01 00:00:00 ┆ 1960-01-01 00:00:00 ┆ 424        │
│ 1959-12-01 00:00:00 ┆ 1960-02-01 00:00:00 ┆ 407        │
│ 1959-12-01 00:00:00 ┆ 1960-03-01 00:00:00 ┆ 471        │
└─────────────────────┴─────────────────────┘────────────┘

Notice the output has three columns: observed_time (the last training timestamp), time (the forecast timestamps), and Passengers (the predicted values).

What we built

You have installed Yohou-Nixtla, loaded a dataset, fit an AutoARIMA forecaster, and generated a 12-month forecast. The same fit/predict pattern works for all 15 forecasters in the package.

Next steps