Refactor to a class about kmeans

[maasencioh]

why TypeScript? I think that the class is not a bad idea, do you see any advantage doing this?

[stropitek]

I think that's a very good idea. People will be familiar with the API. I think most people doing ML have worked with python and scikit before.

@maasencioh we've been experimenting with Typescript and it's awesome.

• Auto-completion and documentation as you code, really useful for ML where you often see have complex types
• Fix typos and type errors before even running the program
• Thinking with types in mind helps you to design a better API IMHO
• It's still just js at the end
• We often have a build step anyway (for example to support es6 modules)
• People have already expressed their desire to see MLJS in typescript

[jobo322]

Syntax is like Java, for reading the code maybe it's better, you need to write more code for the same result but personally I like the comments functionality for the argument types

[lpatiny]

Migrating to a class would be great indeed but I would prefer to have something similar to pca, pls and other packages

const kMeans = new KMeans(dataset, options); (or new KMeans(dataset, numberClusters, options))

const newPoints = [[4.9, 3.2, 1.2, 0.4], [5.4, 3.3, 1.4, 0.9]];
console.log(kMeans.predict(newPoints)); // project new points into the PCA space

const model2 = kMeans.toJSON(); // to save the model

const kMeans2 = KMeans.load(model);
kMeans2.predict(newPoints);