simple_knn

Calculate k-nearest neighbours(knn) with a small subset of lodash with chaining capability

Features

• Calculate knn from List<List<num>> data
• Provide several helper methods to handle List<List<num>> data
  • normalize
  • map
  • sortByColumn
  • zip
  • sum
  • and more
• Simple Unit Testing

Usage

num result = LodashChain.knn(trainingSet, LodashChain.initial(testPoint), k: k)

See example for a accuracy test loop.

import "package:simple_knn/simple_knn.dart";
import 'data.dart';

void main() {
  final testSetSize = 100;
  final List<List<num>> testSet = alldata.sublist(0, testSetSize);
  final List<List<num>> trainingSet = alldata.sublist(0, testSetSize);

  for (int k = 1; k < 7; k++) {
    var successes = testSet.where((List<num> testPoint) =>
        LodashChain.knn(trainingSet, LodashChain.initial(testPoint), k: k) ==
        testPoint.last);
    var accuracy = (successes.length / testSetSize) * 100;
    print("result for k=$k: $accuracy%");
  }
}

Unit Testing

• ✓ constructor and shape two dimensions filled
• ✓ constructor and shape only one dimension filled
• ✓ constructor and shape only scalar
• ✓ slice two dimensions
• ✓ slice only one dimension filled
• ✓ slice only scalar
• ✓ lastInChain two dimensions
• ✓ lastInChain only one dimension filled
• ✓ lastInChain only scalar
• ✓ first two dimensions
• ✓ first only one dimension filled
• ✓ first only scalar
• ✓ sortByColumn two dimensions
• ✓ sortByColumn only one dimension filled
• ✓ sortByColumn only scalar
• ✓ size two dimensions
• ✓ size only one dimension filled
• ✓ size only scalar
• ✓ map two dimensions
• ✓ map only one dimension filled
• ✓ map only scalar
• ✓ filter two dimensions
• ✓ filter only one dimension filled
• ✓ filter only scalar
• ✓ countByToPairs two dimensions
• ✓ countByToPairs only one dimension filled
• ✓ countByToPairs only scalar
• ✓ zip two dimensions
• ✓ zip only one dimension filled
• ✓ zip only scalar
• ✓ sum two dimensions
• ✓ sum only one dimension filled
• ✓ sum only scalar
• ✓ distance chain distance chain
• ✓ cloning two dimensions filled
• ✓ cloning only one dimension filled
• ✓ cloning only scalar
• ✓ normalize two dimensions filled
• ✓ normalize only one dimension filled
• ✓ normalize only scalar
• ✓ static min and max min
• ✓ static min and max max
• ✓ static distance simple v1 and v2
• ✓ static distance simple v1 and v2
• ✓ static initial of array length 4
• ✓ static initial of array length 1
• ✓ knn knn

The implementation is heavily influenced by the Udemy Course "Machine learning with javascript" by Stephen Grider. Thanks a lot to Stephen for his great explanations about knn using the Javascript lodash library.