Consider supporting Typescript

[oveddan]

I want to open the discussion here around allowing models to be developed in typescript in ml5. I feel like using Typescript would bring many advantages.

• We are dealing often with new libraries to import that in general are made in typescript. These libraries have a lot of unknowns; being able to have autosuggest for all of the possible library methods when we don't know the full api would make things much easier.
• We would know right away when we have typos or use a tensorflow.js or third party method incorrectly, saving a lot of debug time.
• When new developers come into contributing ml5, they can more quickly get accustomed to what all the methods are by just pressing a dot when typing in the IDE and having all the options appear right away. If we encourage descriptions on the method signatures, these would show up too. They'd also be less scared of mistyping something and breaking things because typescript lets you know right away if there was a mistake.
• Typescript annotations make reading code a lot easier, especially when there are tons of different internal types going around.

I know there is the argument that we want to make ml5 easy for people new to coding to contribute to, and that by throwing in Typescript they'd have to learn something new on top of javascript and it would potentially hinder them from contributing. I actually would think in the long term it would make it easier for them to contribute and make it faster for them to fix bugs and add features because of the points stated above. They should be able to learn the basics of Typescript by seeing existing code in the repo and reading some intro documentation (it's very well documented) and once they get over that hump with VSCode's awesome typescript integration they'd have a much better experience coding with the repo.

I think a basic start would be to allow (but not require) new models to be made in Typescript. This could be a trial.

What are your thoughts?

[huan]

I just came here and saw this awesome ml5js module for js users, it's really fantastic!

And I would like to vote YES for writing code in TypeScript because it will really help us to save lots of time to maintain the code with higher quality.

Here's a wonderful TypeScript talk from Anders Hejlsberg at here

[cvalenzuela]

We have discussed this a few times, but given the size of the project now, I think this is a very reasonable change to explore and possible make. Will take some time to update the source but might be worth it. Perhaps we can discuss this in more detailed on our January meetup? @shiffman, what are your thoughts on this?

[nathanvogel]

I've never really used Typescript, but when I quickly hacked ml5 to have the classifier support Canvas input, understanding the supported types of the arguments and variables was my main struggle.

[huan]

@nathanvogel Yes, that's exactly why we need TypeScript!

[shiffman]

I am very happy to revisit this discussion! My thinking has been to stay with JavaScript in terms of the project being easier to contribute to for beginner coders and alignment with p5.js (I like opening up the source code in class to show how the library works and have it be recognizable for students). @oveddan brings up a lot of good points. Let's discuss next week during our meetings at ITP, I'll be sending out some e-mails about that soon!

[shiffman]

This came up a bit in our discussions at ITP yesterday. One idea proposed was to do this incrementally and maybe use typescript for new models/classes we are adding to ml5 and see how that goes. Would that be possible? Keeping the existing code base as is (for now) and adding a new class with typescript as a trial run?

[shiffman]

Here's a nice quick intro to TypeScript for those so interested! (Like me)

https://dev.to/robertcoopercode/get-started-with-typescript-in-2019-6hd

[oveddan]

Hey just got back in town. I think trying it for one or two new models makes sense. It should definitely be possible to do with javascript and typescript side by side. I can take a stab at this for the BodyPix implementation.

[mateja176]

Right now when you attempt to install type definitions for ml5 like so

npm install --save-dev @types/ml5 

you are met with the following error:

npm ERR! code E404
npm ERR! 404 Not Found - GET https://registry.npmjs.org/@types%2fml-5 - Not found
npm ERR! 404 
npm ERR! 404  '@types/ml-5@latest' is not in the npm registry.
npm ERR! 404 You should bug the author to publish it (or use the name yourself!)
npm ERR! 404 
npm ERR! 404 Note that you can also install from a
npm ERR! 404 tarball, folder, http url, or git url.

npm ERR! A complete log of this run can be found in:
npm ERR!     /home/username/.npm/_logs/2019-06-18T21_08_22_838Z-debug.log

As a deeply invested typescript user the only thing I lacked was type definitions for the library. However, based on a minimal introductory project, ml5 has held up to, if not surpassed, my expectations. I like it so much it fact ( and the whole concept of a high level, accessible machine learning library ) that I would love to contribute.

Following is a list of helpful resources:

https://www.typescriptlang.org/docs/handbook/declaration-files/publishing.html
https://github.com/DefinitelyTyped/DefinitelyTyped#how-can-i-contribute
http://definitelytyped.org/guides/contributing.html

[dikarel]

Hi, I like this library and I'm in support of having a @types/ml5 type definition library. In my experience, autocomplete has helped me learn quickly as a beginner -- it lets me explore possibilities and without constantly stumbling over trivial mistakes.

Has anyone started working on this? I'm happy to take a quick stab otherwise.
In case this approach conflicts with existing plans, let me know!

EDIT: A quick draft that I did

[kresli]

any updates?

[joeyklee]

@kresli - Thanks for checking in. It definitely is helpful to help gauge interest in certain features.
No updates at the moment, but this is something that is on our minds as a discussion point. If anything changes, we'll be sure to ping this issue!