search

plonerma / understanding-maml

An Interactive Introduction to Model-Agnostic Meta-Learning: Exploring MAML and its extensions
https://plonerma.github.io/understanding-maml/
1 stars 1 forks source link

Comparison of methods #7

Closed plonerma closed 3 years ago

plonerma commented 3 years ago
luis-mueller commented 3 years ago

Im working on them sliders!

plonerma commented 3 years ago

052fdcf35f7290b2207e268f7b8ca9f82b543827

luis-mueller commented 3 years ago

Awesome, so now we just have to discuss what we do with the webapp. Wanna check it out (my latest commit) and tell me what you think? I added a toggle for activating the methods to maybe just compare two at a time or so...

Is the reg. coeff. of iMAML aleady implemented in js? If yes, we should connect the slider. Didn't find the variable.

luis-mueller commented 3 years ago

Btw I suggest we just leave the webapp where it is and close this issue. It's not optimal because it overlaps but I don't know if it is better to put it on a standalone page.

plonerma commented 3 years ago

I will take a look (regarding the regularization parameter).

Yes, the way it is right now should be fine. Plus, this way it has some context (instead of being isolated).

I really like the visual effect of the disabled methods - makes it quite clear what is happening (at least for me, knowing what is about anyways...).

One last thing that we should add: Either their should be a dropdown box for the different tasks or we should make it clear that there is more than one.

plonerma commented 3 years ago

It would probably also be nicer if the sliders had equal length, but I'm not sure if there is an easy way to achieve this using the slider component I wrote...

plonerma commented 3 years ago

Have you noticed, there is a really weird behavior: only MAML will update, if it is solely active, and iMAML only updates, if all methods are active? Maybe there is something off with the filter...

plonerma commented 3 years ago

Ah, if maml is filtered away, all following is reference the wrong alg. Fixed in 6adcd8212bc10ac3180e7cf68e1af58ab6ef3f77

luis-mueller commented 3 years ago

ok awesome, so we keep the figure there for now and close this?

plonerma commented 3 years ago

I will work on this tomorrow:

and then close it.