Exciting stuff!

What browser is this? 👀

That’s awesome! I wonder what the lift is to ship it with the precompiled binary (to remove the Rust dependency on the dep users), like Explorer and friends.

Since I’m interested in using this, happy to have a look at that if you don’t mind.

I think this is a good sign for adoption though. More companies using Elixir means more developers that dislike it will have to use it at work, meaning more dislikes in SO surveys.

Ahhh, modularity *chef’s kiss*

Ohhhh that looks nice!

Is it a library? Or more like a full fledged tool?

That would make sense since there’s no concept of a process in the JS land (yet haha)

This is truly fascinating stuff!

I was reading the example code, and I see the Elixir node it sends the message to a process named “counter” in the JS node, but I couldn’t see where the JS node registers this name. Did I miss something or is the JS code receiving messages directed at all processes?

Yeap, might be just crazy enough to actually work haha

What if you register a custom reporter that stores the test states, and then query it from the callback?

It is a bit hacky, but should work.

The unknown here for me is whether the reporter gets notified before or after the callbacks are run.

That is clean!

I’m going to take some inspiration from yours and add a “copied” tooltip and a “loading” state.

I was building some Livebook apps last Friday and really missed a “copy to clipboard” button there.

I love his stuff!

That is true for factors, not for sums. Like, using primes I can be sure the products will be unique, but I cannot ensure the sum of products will be unique. In essence, you can have more than one set of 4 products that yield the same sum, which can cause you to overcount.

Got it! In that case I posted a solution using it earlier, and I also link to this great video by 3blue1brown about convolutions. It is great to build that intuition about what is actually happening there.

I don’t want to spoil anything, but you should have a look at a thread I posted earlier today once you’re done with the puzzle 😄

Similar logic applies to your kernel, except the two required dimensions are output channels and input channels. The latter has to match the second dimension of your input. So for the same case you have a {n_kernels, 1, x, y} dimensional tensor kernel.

It is supported, but you need two dimensions for batch size and channels and then n dimensions for your n-dimensional input.

So if you need to run a convolution over a 2D grid, and you don’t care about channels or batches, you need a {1, 1, x, y} tensor.