rs86

Just don't punish yourself over it. There is more information than anyone could handle. It's really a matter of realizing failure is a part of life. You can not avoid failure, but you can sure handle it.

How could one add undertesting to the model?

I don't think those kinds of exercises will teach you about functional programming... Here's why. You can understand imperative programming as a sequence of operations. But functional programming is more like a composition of transformations. It really requires study because you need to think differently.

I would suggest rewriting the standard library of a functional language.... Read the standard functions on lists, maps, tuples, folds, maps, and reimplement them...

I would ask you questions about Haskell and functional programming, if you get them right it will definitely be a plus, but if you say you know Haskell and can't write basic monads from scratch you are doomed

Such naivete

Can't the module system be used for ad hoc polymorphism?

Elm is quite different from purescript or Haskell... It shares syntax and the purity semantics but its type system is much less expressive (no type classes); no do notation; no monads...

Elm has managed effects and purity, and I had immense satisfaction in using it... Managed effects means that doing http requests and working with DOM events are managed from a runtime and from the point of view of the programmer they are just returning values that describe the effectful actions they want to perform, and any changes to programming state from those actions are apparently as function arguments...

The language has its warts, but I would use it if it were not for the poor project management and support.

This is ridiculously full of unfounded statements, naivety and unsound reasoning

It is a subject of math. Safety and other qualifies of it can be analysed logically... Or not, as the article points out

Theorems are strings of symbols and can thus be encoded as numbers, and statements about numbers become statements about theorems. That's why the theorem applies to systems at least as strong as arithmetics

This is about the possibility of proving certain things about the model, not about properties of any implementation. Properties include for example the relation between the amount and quality of data available for training and the accuracy of the model. If a model is proven to learn too slowly it costs more to train it. And some models might never approximate a target function.

I have mixed feelings about using Elixir (or Erlang); as far as I understand the platform, it is about building fault tolerant systems/high availability, specially in the presence of hardware failure. I think those are handled well by cloud service providers; they didn't exist during the 80s.

I think performance is better compared to Ruby and Python, but then again my experience with web applications is that the domains are best modeled using classes.

For writing networking code and protocols, the binary pattern matching is amazing, though. The Plug libraries are a pleasure to use also.