sklearn-progestimator

A scikit-learn estimator which wraps a base estimator to provide facilities for time series problems where previous predictions are used as features.

---

My other goal here is to use this as a toy project for learning standard tools for Python development (pytest, tox, setuptools, pdb)

Installation

Using pip:

pip install scikit-learn-progestimator

Using conda:

WIP

Usage

When calling estimator.fit(X,y), y with time lag 1 is appended to X before fitting the model. When calling estimator.predict(X), for each sample in X, the prediction uses the previous known value for y (either true or predicted) as an additional feature.

This wrapper implements the standard estimator API. As such, it should play well with the rest of scikit-learn.

>>> from sklearn.linear_model import LinearRegression
    from progestimator.prog_regression import ProgressiveRegression
    y = np.array([[1.0], [3.0], [4.0], [7.0], [15.0], [31.0]])
    X = np.ones(([1.0], [1.0], [1.0], [1.0], [1.0], [1.0]])
    model = ProgressiveRegression(LinearRegression()) 
    model.fit(X,y)

>>> model.predict(([1.0], [1.0], [1.0], [1.0], [1.0], [1.0]]))

array([[  64.98224852],
       [ 137.08896047],
       [ 290.09172322],
       [ 614.74728963],
       [1303.63182285],
       [2765.37143003]])

Development

Environment

Create the development environment and activate it

conda env create -f environment.yml
conda activate progregressor-dev

Alternatively, the environment can be replicated with pip and requirements.txt

Tests

In tests

python -m pytest