Introduction

Gumbi simplifies the steps needed to build a Gaussian Process model from tabular data. It takes care of shaping, transforming, and standardizing data as necessary while applying best practices and sensible defaults to the construction of the GP model itself. Taking inspiration from popular packages such as Bambi and Seaborn, Gumbi’s aim is to allow quick iteration on both model structure and prediction visualization. Gumbi is primarily built on top of Pymc, though additional support for GPflow is planned.

Quickstart

Read in some data and store it as a Gumbi DataSet:

[1]:
import gumbi as gmb
import seaborn as sns
cars = sns.load_dataset('mpg').dropna()
ds = gmb.DataSet(cars, outputs=['mpg', 'acceleration'], log_vars=['mpg', 'acceleration', 'horsepower'])
WARNING (pytensor.tensor.blas): Using NumPy C-API based implementation for BLAS functions.

Create a Gumbi GP object and fit a model that predicts mpg from horsepower:

[2]:
gp = gmb.GP(ds)
gp.fit(outputs=['mpg'], continuous_dims=['horsepower']);

Make predictions and plot!

[3]:
X = gp.prepare_grid()
y = gp.predict_grid()
gmb.ParrayPlotter(X, y).plot()
[3]:
<Axes: xlabel='horsepower', ylabel='mpg'>
../../_images/notebooks_getting_started_introduction_7_1.png

More complex GPs are also possible, such as correlated multi-input and multi-output systems, demonstrated in the example notebooks.

Installation

Via pip

pip install gumbi

Bleeding edge

pip install git+git://github.com/JohnGoertz/Gumbi.git@develop

Local development

  • Clone the repo and navigate to the new directory

    • git clone https://gitlab.com/JohnGoertz/gumbi gumbi

    • cd gumbi

  • Create a new conda environment using mamba

    • conda install mamba

    • mamba install -f gumbi_env.yaml

  • Install gumbi via pip in editable/development mode

    • From within the gumbi repo

    • pip install --editable ./

  • To update the gumbi module

    • From within the gumbi repo

    • git pull