justinlovinger / optimal-py-ill

Ensemble that enables supervised learning algorithms to effectively learn incrementally

Infinite Lattice Learner (ILL)

An ensemble learning algorithm that enables supervised learning algorithms to effectively learn incrementally. Introduced in the paper, Infinite Lattice Learner: an ensemble for incremental learning.

Although the algorithm itself can be very efficient, this code was developed primarily for research and is not as efficient as it could be.

Installation

Nix

Add to Nix script like

...
let
  ill = (import (pkgs.fetchFromGitHub {
    owner = "JustinLovinger";
    repo = "ill";
    rev = "LATEST_VERSION_TAG";
    sha256 = "SHA25_OF_LATEST_VERSION_TAG";
  }) { inherit pkgs; });
in pkgs.python2Packages.buildPythonPackage {
  ...
  propagatedBuildInputs = [ ill ];
  ...
}

Get sha256 with nix run nixpkgs.nix-prefetch-github -c nix-prefetch-github --rev "LATEST_VERSION_TAG" JustinLovinger ill.

Other

Obtain dependencies listed in default.nix propagatedBuildInputs, and copy ill.py to your project.

Example Usage

import functools

from learning import datasets, validation, LinearRegressionModel
from ill import ILL, get_neighborhood_k_nearest

# Grab the popular iris dataset, from 'learning'
dataset = datasets.get_iris()

# Make an underlying model for ILL
# See 'learning' for more details
underlying_model = LinearRegressionModel(4, 3)

# Make an ILL ensemble of our underlying model
# See code for more options
model = ILL(
    underlying_model,
    grid_spacing=0.5,
    neighborhood_func=functools.partial(
        get_neighborhood_k_nearest, k_nearest=5))

# Lets train our ILL
# First, we'll split our dataset into training and testing sets
# Our training set will contain 30 samples from each class
training_set, testing_set = validation.make_train_test_sets(
    *dataset, train_per_class=30)

# We could customize training and stopping criteria through
# the arguments of train, but the defaults should be sufficient here
model.train(*training_set)

# Our ILL should converge in a few moments
# Lets see how our ILL does on the testing set
print 'Testing accuracy:', validation.get_accuracy(model, *testing_set)

Development

Enter a Nix shell with nix-shell.