add cross-validation feature

[sanjayaksaxena]

No description provided.

[sanjayaksaxena]

Use wink-helpers cross validation instead!

[jtara1]

not sure if this is proper implementation, but I implemented k folds cross validation func

const _ = require('lodash');
const winkHelpers = require('wink-helpers');

/**
 * zip data & labels, shuffle them, partition them into the number of folds
 * specified
 * @param data {[]}
 * @param labels {[]}
 * @param folds {number}
 * @returns {Array<Array<Array<*>>>} an array of groups of zipped data & labels
 */
module.exports = function kFoldsCrossValidation(data, labels, folds) {
	if (data.length < 2) throw new Error('only 1 row for data given');

	let dataAndLabels = _.zip(data, labels);
	dataAndLabels = winkHelpers.array.shuffle(dataAndLabels);

	return partition(dataAndLabels, folds);
};

function partition(dataAndLabels, numberOfGroups) {
	const groups = [];
	const len = dataAndLabels.length;
	const partitionSize = Math.ceil(len / numberOfGroups);

	for (let i = 0; i < len; i += partitionSize) {
		groups.push(dataAndLabels.slice(i, i + partitionSize));
	}

	return groups;
}

// example usage

// made up data & labels
const start = 100;
const end = 128;
const diffSplit = Math.floor((end - start) / 2);

let a = _.range(start, end);
const labels1 = new Array(diffSplit);
const labels2 = new Array(diffSplit);
labels1.fill('a');
labels2.fill('b');

let l = _.concat(labels1, labels2);

const groups = module.exports(a, l, 3);
console.log(groups);

// test group
let [data, labels] = _.unzip(groups[0]);
console.log(data);
console.log(labels);

or just use https://visualml.io/jsmlt/docs/function/index.html#static-function-trainTestSplit

[sanjayaksaxena]

Hi @jtara1

Thanks a lot for your inputs!

In fact wink-helpers have an API for cross validation, which is not yet completely documented. It lets you compute macro-averaged avgPrecision, avgRecall, and avgFMeasure apart from many more details.

You may like to review the same.

We would keep your inputs in mind at the time of completing all of these.

Best,
Sanjaya

[jtara1]

I was a little confused because it doesn't do the part of the cross validation where you partition data into testing and training sets. https://en.wikipedia.org/wiki/Cross-validation_(statistics)

I had another question about this module, I'll just open another issue.