Hacker News Reader

Comment by ptype | Hacker News Reader

I love Python. The language is a joy, the eco system is fantastic. But yes, let’s be honest, if you can not vectorise your code it is slow, and I think that will be its downfall eventually.

I’m excited about Julia, I hope it gains popularity and the eco system grows. Until then, and in particular until the data frames story can compete with pandas, it Python with Cython for me, but I’d rather skip the Cython if it was not necessary for performance.

Any early adopters running Julia in production with stories to share?

View on HN

I feel bit ambivalent about Python as it's a nice language for prototyping and quickly hacking things done. Yet I'm always baffled when I read Numpy's or Matplotlib's documentation and try to make sense of it as it can be (or at least feel) so complex and highly ambiguous. Eg. sometimes there is no/very brief examples at Numpy's documentation pages how the method works and most results from Google are only about advanced implementations, not about the basics of the method itself. In Matplotlib I still don't understand what is the right way of initializing a pyplot, there seems to be a million ways to do it and a million parameters you can give. API changes and inconsistencies too pain me at times (Pandas comes to my mind). While not a fault of Python as a language I think they greatly contribute to the experience of using Python.

Also I don't feel like the culture of Python programming focuses too much on documenting things which makes reading code at times like transcribing ancient Latin manuscripts. Maybe a good analogy would be JS back in the days with global jQuery scripts. Too unrestricted and free-form maybe. I'd wish Python became more like Kotlin with very clear patterns and great IDE support (in addition to PyCharm).

Well those are at least my experiences and feel free to disagree with me.

> In Matplotlib I still don't understand what is the right way of initializing a pyplot, there seems to be a million ways to do it and a million parameters you can give. API changes and inconsistencies too pain me at times

Matplotlib has the worst API of all Python libraries I have used over the years!

If there were a fork of it that got rid of Matlab-way of doing things (keeping only OOP style) and with consistent names (no more `twowords` and `two_words`), I would gladly switch in a heartbeat.

This is maybe more hearsay, I only briefly tried to use Julia.

It is true, Julia has great features for performant code. It is, however, focussing too much on being a matlab competitor in my opinion. It will not be a language that you use to write a "normal" (i.e. non-numeric or CRUD) dynamic website in. My general observation however is, that you need to attract this crow, if you want to have an ecosystem with a variety of tooling. And it is the neat thing about Python (and Haskell).

I'm using Julia and loving it. I've built a bunch of differential equation solvers which routinely outperform the classic C++/Fortran codes. I started out without "software development" experience but Julia and its community got me up to speed and helped me build something quite unique. Now Julia is the only language that has the numerical libraries I need to do my research.

In fact, the whole library story in Python/MATLAB is quite overblown. If you're doing something which is actually new, like PhD methods research, you need to be writing a lot of stuff from scratch. And in that case, you usually cannot get by with vectorizing everything... and vectorization always has the issue with temporary arrays too. Meanwhile, Julia's type system makes everything fast (which is a plus when trying to publish a paper on it!) but also get cool extra features for free like GPU support and arbitrary precision. For people developing and testing new methods, Julia is the best tool right now.

>If you're doing something which is actually new, like PhD methods research, you need to be writing a lot of stuff from scratch.

Sometimes you also reuse a lot of stuff, it depends. For machine learning in particular, almost everything uses some sort of gradient-based optimisation algorithm. In these cases, it is very useful to have an automatic differentiation library. My coworker said Julia has about 3, IIRC, and until an amalgamation of them is merged into the standard library Julia isn't completely ready.

> until an amalgamation of them is merged into the standard library Julia isn't completely ready

That doesn't quite make sense. I'm sure there are a few more autodiff libraries in Julia than 3. You would just use the one that fits your use case.

PS. Ohh re-reading your comment, you want an autodiff library in Julia's standard library. That is very unlikely to happen since it's not (very) hard to cook up an autodiff library and autodiff is not widely used. Julia is not like Matlab, where you have to have everything in the standard library.

Yeah, you read it correctly the second time :)

>That is very unlikely to happen

Well, my co-worker says people are working on it. I asked him whether I should try to adopt Julia and "not until autodiff is in the standard library" was his response.

>autodiff is not widely used

If you look only at the machine learning community, that is false. Autodiff is used all the time, to optimise loss functions using a variant of gradient descent. For neural nets, gaussian processes, SVMs... Not decision trees/forests, but these are not more popular than all the others combined.

Maybe there was some small miscommunication here – we are indeed working on "one AD to rule them all" (Capstan, [1]), but it won't go in the standard library as such (there's no need as it will be just as good as a package).

That said, Capstan relies on very new compiler technology and still requires some deep changes there, so will only work on future Julia versions (which may be what he was referring to).

Until then, Flux [2] has an AD that's well-suited to ML and works on current Julia versions.

[1] https://github.com/JuliaDiff/Capstan.jl/ [2] https://fluxml.github.io/

Reply on news.ycombinator.com