[DigitalTerminal]

Sounds spiffy. Also, I must find time to learn Julia. One problem with these 'out of a box' cookbook systems is that you only know approximately what you are doing. If it works, that may be fine, but for statistical systems the outcome depends heavily on the formulation, is strongly influence by random or irrational correlations and is more art than science. If you remove the element of experience that animates the "art", you get nonsense. Big box programming tools can make it hard to know what correlations are being established, and this is true with even simple neural networks. By the time they have many layers and filters in them, all hope is fled. You buy the result or you don't: simple as that.

[xiphias2]

I have nowhere experienced more looking under the hood feeling then when I am using Julia.

The source code of the packages sit on my computer, I can (and sometimes do) modify it, and I can still compile the same function to SIMD or GPU.

[GorgeRonde]

Although I have never used Julia, this is something that is possible in Ruby, and I think this feature is seriously underrated – i.e. to be able to jump quickly to the definition of external code, modify it and run it. When you work with multiple libraries/repo this is very valuable when debugging, saves a lot of time.

It should also be possible with Node.js I think (node_modules).

[xiphias2]

Don't get me wrong, I love Ruby, and my code was originally in Ruby, but the 20x speedup and better math libraries in Julia without sacrificing the speed of development too much made the switch for me very easy.

Actually translating Ruby code to Julia was much easier than I thought (the only real difference is the indexing).

The 1 based indexing really sucks in the PTX assembly output of Julia as well, I see a lot of useless increment and decrement operations when I don't expect it.