-
Notifications
You must be signed in to change notification settings - Fork 29
Description
I am struggling to understand the point of offering something so poorly maintained as MLJ.
I will try to be positive and offer constructive criticism.
There are so many issues that it has hard to figure where to start. But I will try:
-
Documentation is missing, incorrect, outdated. There are several cases of links to non existing pages, such as https://alan-turing-institute.github.io/MLJ.jl/dev/loading_model_code/#Loading-Model-Code
-
Code shown in the model browser pages often fails. For example, the package BetaML, which is shown in one of the tutorials, does not work.
-
I use Go quite extensively, so I was hoping that Julia, also a modern language, would compare well to GO. Unfortunately that is not the case. When I compare the practice of development and testing in Go versus Julia, it is clear that testing in MLJ is not a primary consideration. There is "test" subdirectory under the MLJ, but the code is so incomplete....
-
And then, there is the issue of debugging, which is not specific to MLJ, but, debugging in Julia is so slow that it is simply useless.
-
I really was hoping to like Julia and use it for serious data science work. Unfortunately working in Julia is more of a pain than a joy, to the extent that I could not use it for the reasons already expressed, and I will recommend my colleagues to stay away.