AutoAI-TS

An opens-source implementation of the AutoAI-TS model selection framework with minimal dependencies.

AutoAI-TS is a zero-config model selection framework for time series. It performs data validation, automatic look-back computation, and model selection with the T-Daub algorithm.

For more details on AutoAI-TS, read the original paper: AutoAI-TS: AutoAI for Time Series Forecasting.

Installation

This project uses Python 3.13. We recommend you install the conda package manager to install the correct Python version and packages.

The model selection implementation only depends on numpy and pandas. In most cases, your ML project should have these libraries already installed.

There are two version of the environment: standard and minimal.

Standard Installation (Recommended)

The standard environment installs the model selection dependencies and the following ML libraries: scikit-learn, statsmodels, and xgboost.

conda env create -f environment.yml
conda activate autoai-ts

Minimal Installation

If you want to use your own set of models, you can install the minimal environment. This will only install numpy and pandas:

conda env create -f minimal-environment.yml
conda activate autoai-ts

Example

With the training data as X and your targets as y. The data can be a pandas DataFrame or a Numpy array.

from autoai_ts import AutoAITS
from sklearn import linear_model, svm, ensemble

# you need a set of models to use for model selection
pipelines = [linear_model.LinearRegression(), svm.SVR(), ensemble.RandomForestRegressor()]

model = AutoAITS(pipelines)
model.fit(X, y)

model.score(X, y)

If you are using the standard environment, a set of pipeline is available through the pipeline.py module. dataset.py contains some time series dataset that you can use to test AutoAI-TS. Here is a more complete example on the Passenger flights dataset with our set of pipelines:

from autoai_ts import AutoAITS
from pipeline import create_pipelines
from sklearn.model_selection import train_test_split
import dataset

### Passenger flights
df = dataset.get_flights_dataset()
X, y = dataset.to_supervised(df)

# train-test split 80/20; not shuffled because time series
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, shuffle=False)

pipelines = create_pipelines()

model = AutoAITS(pipelines, positive_idx=[2])
model.fit(X_train, y_train)

scores = model.score(X_test, y_test)
print("Evaluation: ", scores)

License

Licensed under the GNU AGPLv3 license.