Add k-nearest neighbors algorithm.

This commit is contained in:
Oleksii Trekhleb 2020-12-16 08:07:08 +01:00
parent b13291df62
commit 4623bb906f
7 changed files with 190 additions and 126 deletions

View File

@ -143,7 +143,7 @@ a set of rules that precisely define a sequence of operations.
* `B` [Caesar Cipher](src/algorithms/cryptography/caesar-cipher) - simple substitution cipher * `B` [Caesar Cipher](src/algorithms/cryptography/caesar-cipher) - simple substitution cipher
* **Machine Learning** * **Machine Learning**
* `B` [NanoNeuron](https://github.com/trekhleb/nano-neuron) - 7 simple JS functions that illustrate how machines can actually learn (forward/backward propagation) * `B` [NanoNeuron](https://github.com/trekhleb/nano-neuron) - 7 simple JS functions that illustrate how machines can actually learn (forward/backward propagation)
* `B` [KNN](src/algorithms/ML/KNN) - K Nearest Neighbors * `B` [k-NN](src/algorithms/ml/knn) - k-nearest neighbors classification algorithm
* **Uncategorized** * **Uncategorized**
* `B` [Tower of Hanoi](src/algorithms/uncategorized/hanoi-tower) * `B` [Tower of Hanoi](src/algorithms/uncategorized/hanoi-tower)
* `B` [Square Matrix Rotation](src/algorithms/uncategorized/square-matrix-rotation) - in-place algorithm * `B` [Square Matrix Rotation](src/algorithms/uncategorized/square-matrix-rotation) - in-place algorithm

View File

@ -1,23 +0,0 @@
# KNN Algorithm
KNN stands for K Nearest Neighbors. KNN is a supervised Machine Learning algorithm. It's a classification algorithm, determining the class of a sample vector using a sample data.
The idea is to calculate the similarity between two data points on the basis of a distance metric. Euclidean distance is used mostly for this task. The algorithm is as follows -
1. Check for errors like invalid data/labels.
2. Calculate the euclidean distance of all the data points in training data with the classification point
3. Sort the distances of points along with their classes in ascending order
4. Take the initial "K" classes and find the mode to get the most similar class
5. Report the most similar class
Here is a visualization for better understanding -
![KNN Visualization](https://media.geeksforgeeks.org/wp-content/uploads/graph2-2.png)
Here, as we can see, the classification of unknown points will be judged by their proximity to other points.
It is important to note that "K" is preferred to have odd values in order to break ties. Usually "K" is taken as 3 or 5.
## References
- [GeeksforGeeks](https://media.geeksforgeeks.org/wp-content/uploads/graph2-2.png)

View File

@ -1,42 +0,0 @@
import KNN from '../knn';
describe('KNN', () => {
test('should throw an error on invalid data', () => {
expect(() => {
KNN();
}).toThrowError();
});
test('should throw an error on invalid labels', () => {
const nolabels = () => {
KNN([[1, 1]]);
};
expect(nolabels).toThrowError();
});
it('should throw an error on not giving classification vector', () => {
const noclassification = () => {
KNN([[1, 1]], [1]);
};
expect(noclassification).toThrowError();
});
it('should throw an error on not giving classification vector', () => {
const inconsistent = () => {
KNN([[1, 1]], [1], [1]);
};
expect(inconsistent).toThrowError();
});
it('should find the nearest neighbour', () => {
let dataX = [[1, 1], [2, 2]];
let dataY = [1, 2];
expect(KNN(dataX, dataY, [1, 1])).toBe(1);
dataX = [[1, 1], [6, 2], [3, 3], [4, 5], [9, 2], [2, 4], [8, 7]];
dataY = [1, 2, 1, 2, 1, 2, 1];
expect(KNN(dataX, dataY, [1.25, 1.25]))
.toBe(1);
dataX = [[1, 1], [6, 2], [3, 3], [4, 5], [9, 2], [2, 4], [8, 7]];
dataY = [1, 2, 1, 2, 1, 2, 1];
expect(KNN(dataX, dataY, [1.25, 1.25], 5))
.toBe(2);
});
});

View File

@ -1,60 +0,0 @@
/**
* @param {object} dataY
* @param {object} dataX
* @param {object} toClassify
* @param {number} k
* @return {number}
*/
export default function KNN(dataX, dataY, toClassify, K) {
let k = -1;
if (K === undefined) {
k = 3;
} else {
k = K;
}
// creating function to calculate the euclidean distance between 2 vectors
function euclideanDistance(x1, x2) {
// checking errors
if (x1.length !== x2.length) {
throw new Error('inconsistency between data and classification vector.');
}
// calculate the euclidean distance between 2 vectors and return
let totalSSE = 0;
for (let j = 0; j < x1.length; j += 1) {
totalSSE += (x1[j] - x2[j]) ** 2;
}
return Number(Math.sqrt(totalSSE).toFixed(2));
}
// starting algorithm
// calculate distance from toClassify to each point for all dimensions in dataX
// store distance and point's class_index into distance_class_list
let distanceList = [];
for (let i = 0; i < dataX.length; i += 1) {
const tmStore = [];
tmStore.push(euclideanDistance(dataX[i], toClassify));
tmStore.push(dataY[i]);
distanceList[i] = tmStore;
}
// sort distanceList
// take initial k values, count with class index
distanceList = distanceList.sort().slice(0, k);
// count the number of instances of each class in top k members
// with that maintain record of highest count class simultanously
const modeK = {};
const maxm = [-1, -1];
for (let i = 0; i < Math.min(k, distanceList.length); i += 1) {
if (distanceList[i][1] in modeK) modeK[distanceList[i][1]] += 1;
else modeK[distanceList[i][1]] = 1;
if (modeK[distanceList[i][1]] > maxm[0]) {
[maxm[0], maxm[1]] = [modeK[distanceList[i][1]], distanceList[i][1]];
}
}
// return the class with highest count from maxm
return maxm[1];
}

View File

@ -0,0 +1,41 @@
# k-Nearest Neighbors Algorithm
The **k-nearest neighbors algorithm (k-NN)** is a supervised Machine Learning algorithm. It's a classification algorithm, determining the class of a sample vector using a sample data.
In k-NN classification, the output is a class membership. An object is classified by a plurality vote of its neighbors, with the object being assigned to the class most common among its `k` nearest neighbors (`k` is a positive integer, typically small). If `k = 1`, then the object is simply assigned to the class of that single nearest neighbor.
The idea is to calculate the similarity between two data points on the basis of a distance metric. [Euclidean distance](https://en.wikipedia.org/wiki/Euclidean_distance) is used mostly for this task.
![Euclidean distance between two points](https://upload.wikimedia.org/wikipedia/commons/5/55/Euclidean_distance_2d.svg)
_Image source: [Wikipedia](https://en.wikipedia.org/wiki/Euclidean_distance)_
The algorithm is as follows:
1. Check for errors like invalid data/labels.
2. Calculate the euclidean distance of all the data points in training data with the classification point
3. Sort the distances of points along with their classes in ascending order
4. Take the initial `K` classes and find the mode to get the most similar class
5. Report the most similar class
Here is a visualization of k-NN classification for better understanding:
![KNN Visualization 1](https://upload.wikimedia.org/wikipedia/commons/e/e7/KnnClassification.svg)
_Image source: [Wikipedia](https://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm)_
The test sample (green dot) should be classified either to blue squares or to red triangles. If `k = 3` (solid line circle) it is assigned to the red triangles because there are `2` triangles and only `1` square inside the inner circle. If `k = 5` (dashed line circle) it is assigned to the blue squares (`3` squares vs. `2` triangles inside the outer circle).
Another k-NN classification example:
![KNN Visualization 2](https://media.geeksforgeeks.org/wp-content/uploads/graph2-2.png)
_Image source: [GeeksForGeeks](https://media.geeksforgeeks.org/wp-content/uploads/graph2-2.png)_
Here, as we can see, the classification of unknown points will be judged by their proximity to other points.
It is important to note that `K` is preferred to have odd values in order to break ties. Usually `K` is taken as `3` or `5`.
## References
- [k-nearest neighbors algorithm on Wikipedia](https://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm)

View File

@ -0,0 +1,71 @@
import kNN from '../kNN';
describe('kNN', () => {
it('should throw an error on invalid data', () => {
expect(() => {
kNN();
}).toThrowError('Either dataSet or labels or toClassify were not set');
});
it('should throw an error on invalid labels', () => {
const noLabels = () => {
kNN([[1, 1]]);
};
expect(noLabels).toThrowError('Either dataSet or labels or toClassify were not set');
});
it('should throw an error on not giving classification vector', () => {
const noClassification = () => {
kNN([[1, 1]], [1]);
};
expect(noClassification).toThrowError('Either dataSet or labels or toClassify were not set');
});
it('should throw an error on not giving classification vector', () => {
const inconsistent = () => {
kNN([[1, 1]], [1], [1]);
};
expect(inconsistent).toThrowError('Inconsistent vector lengths');
});
it('should find the nearest neighbour', () => {
let dataSet;
let labels;
let toClassify;
let expectedClass;
dataSet = [[1, 1], [2, 2]];
labels = [1, 2];
toClassify = [1, 1];
expectedClass = 1;
expect(kNN(dataSet, labels, toClassify)).toBe(expectedClass);
dataSet = [[1, 1], [6, 2], [3, 3], [4, 5], [9, 2], [2, 4], [8, 7]];
labels = [1, 2, 1, 2, 1, 2, 1];
toClassify = [1.25, 1.25];
expectedClass = 1;
expect(kNN(dataSet, labels, toClassify)).toBe(expectedClass);
dataSet = [[1, 1], [6, 2], [3, 3], [4, 5], [9, 2], [2, 4], [8, 7]];
labels = [1, 2, 1, 2, 1, 2, 1];
toClassify = [1.25, 1.25];
expectedClass = 2;
expect(kNN(dataSet, labels, toClassify, 5)).toBe(expectedClass);
});
it('should find the nearest neighbour with equal distances', () => {
const dataSet = [[0, 0], [1, 1], [0, 2]];
const labels = [1, 3, 3];
const toClassify = [0, 1];
const expectedClass = 3;
expect(kNN(dataSet, labels, toClassify)).toBe(expectedClass);
});
it('should find the nearest neighbour in 3D space', () => {
const dataSet = [[0, 0, 0], [0, 1, 1], [0, 0, 2]];
const labels = [1, 3, 3];
const toClassify = [0, 0, 1];
const expectedClass = 3;
expect(kNN(dataSet, labels, toClassify)).toBe(expectedClass);
});
});

View File

@ -0,0 +1,77 @@
/**
* Calculates calculate the euclidean distance between 2 vectors.
*
* @param {number[]} x1
* @param {number[]} x2
* @returns {number}
*/
function euclideanDistance(x1, x2) {
// Checking for errors.
if (x1.length !== x2.length) {
throw new Error('Inconsistent vector lengths');
}
// Calculate the euclidean distance between 2 vectors and return.
let squaresTotal = 0;
for (let i = 0; i < x1.length; i += 1) {
squaresTotal += (x1[i] - x2[i]) ** 2;
}
return Number(Math.sqrt(squaresTotal).toFixed(2));
}
/**
* Classifies the point in space based on k-nearest neighbors algorithm.
*
* @param {number[][]} dataSet - array of data points, i.e. [[0, 1], [3, 4], [5, 7]]
* @param {number[]} labels - array of classes (labels), i.e. [1, 1, 2]
* @param {number[]} toClassify - the point in space that needs to be classified, i.e. [5, 4]
* @param {number} k - number of nearest neighbors which will be taken into account (preferably odd)
* @return {number} - the class of the point
*/
export default function kNN(
dataSet,
labels,
toClassify,
k = 3,
) {
if (!dataSet || !labels || !toClassify) {
throw new Error('Either dataSet or labels or toClassify were not set');
}
// Calculate distance from toClassify to each point for all dimensions in dataSet.
// Store distance and point's label into distances list.
const distances = [];
for (let i = 0; i < dataSet.length; i += 1) {
distances.push({
dist: euclideanDistance(dataSet[i], toClassify),
label: labels[i],
});
}
// Sort distances list (from closer point to further ones).
// Take initial k values, count with class index
const kNearest = distances.sort((a, b) => {
if (a.dist === b.dist) {
return 0;
}
return a.dist < b.dist ? -1 : 1;
}).slice(0, k);
// Count the number of instances of each class in top k members.
const labelsCounter = {};
let topClass = 0;
let topClassCount = 0;
for (let i = 0; i < kNearest.length; i += 1) {
if (kNearest[i].label in labelsCounter) {
labelsCounter[kNearest[i].label] += 1;
} else {
labelsCounter[kNearest[i].label] = 1;
}
if (labelsCounter[kNearest[i].label] > topClassCount) {
topClassCount = labelsCounter[kNearest[i].label];
topClass = kNearest[i].label;
}
}
// Return the class with highest count.
return topClass;
}