Crate random_world

The random-world crate.

A crate implementing Machine Learning ML methods for confident prediction (e.g., Conformal Predictors) and related ones introduced in the book Algorithmic Learning in a Random World (ALRW).

Goals

• Fast implementation of methods introduced in the book ALRW.
• Should easily allow to wrap existing rust implementations of ML classifiers/scorers.
• (Maybe) allow interfacing to Python.
• (Maybe) can be called as a binary.

Examples

Create a Conformal Predictor with k-NN nonconformity measure, k=2, and with significance level epsilon=0.3, train it on some training set and use it to predict two test vector inputs.

The output predictions will be a matrix, one row per each training input, and one column per label, where each bool element is true if the label conforms the distribution, false otherwise.

#[macro_use(array)]
extern crate ndarray;
extern crate random_world;

use random_world::cp::*;
use random_world::ncm::*;

let ncm = KNN::new(2);
let n_labels = 2;
let mut cp = CP::new(ncm, n_labels, Some(0.3));
let train_inputs = array![[0., 0.],
                          [1., 0.],
                          [0., 1.],
                          [1., 1.],
                          [2., 2.],
                          [1., 2.]];
let train_targets = array![0, 0, 0, 1, 1, 1];
let test_inputs = array![[2., 1.],
                         [2., 2.]];

// Train and predict
cp.train(&train_inputs.view(), &train_targets.view())
  .expect("Failed prediction");
let preds = cp.predict(&test_inputs.view())
              .expect("Failed to predict");
assert!(preds == array![[false, true],
                        [false, true]]);

More examples on deterministic/smooth Conformal Predictors at CP.

Modules

cp	Module defining Conformal Predictors.
exchangeability	Defines Exchangeability Martingales.
ncm	Module defining nonconformity measures.
utils	Utility routines for loading and storing data into files.