libpython-clj VS dtype-next

It’s also worth noting that Clojure has libpython-clj (https://github.com/clj-python/libpython-clj) which offers an interface with Python from another lisp. Here are some advanced ML and dataviz examples using that lib: https://github.com/gigasquid/libpython-clj-examples.

Basically, you take a programming language and make it work on a platform that meant to be programmed using a different PL. Clojure is hosted by design - it's not Java, but can be used to program for JVM. It ain't Javascript, but can be used to target nodejs and browser; not an [official] CLR language, but you can write .Net programs. You can use Clojure to make Flutter apps with ClojureDart. You can integrate Python into Clojure with libpython-clj. Or write Clojure to target Erlang/OTP; or Rust; or R; There's even a clojure-like language for Lua - Fennel.

There's something about Clojure people like so much, they want it to work atop any platform.

https://github.com/Tensegritics/ClojureDart

https://github.com/clj-python/libpython-clj

https://github.com/clojerl/clojerl

https://github.com/clojure-rs/ClojureRS

https://github.com/scicloj/clojisr

https://fennel-lang.org

There's this project that's used a lot for taking advantage of Pythons ecosystem through Clojure JVM. https://github.com/clj-python/libpython-clj

Ahh, makes sense. Well, if you ever wanna steal some of python's thunder, libpython-clj worked great for me lol. Supposedly py4cl fills a similar role in Common Lisp.

High quality interop with any python library via libpython-clj including, but not limited to, keras, numpy, matplotlib, and pandas. This includes zero copy paths from many of those.

libpython-clj now has Java bindings...go tell all your Java friends!

Related (?): https://github.com/cnuernber/dtype-next/blob/master/test/tec...

Tablecloth by generateme is a friendly & expressive table-processing library built on top of tech.ml.dataset & dtype-next, Chris Nuernberger's high-performance data libraries.

There is one other detail here that I found out w/r/t arrays - Clojure's aset implementation returns the previous value; it isn't a faithful wrapper of the JVM's array set value instruction. Due to this if you are using aset on primitive arrays you end up boxing every value you are setting which at least in my tests leads to a performance disadvantage when compared to a tight loop using Java. This is why I have a specialized class implementing an aset that returns void.

If this is something common and the work is numeric, I would highly recommend exploring dtype-next buffer abstraction and tensors. The tensor api supports a nice APL-like substrate for working in index space without having to have the underlying storage "be" a boxed datastructure. You also get the option of off-heap / native tensors that can be zero-copy shuttled between other runtimes.

dtype-next - Major discoverability upgrades for the tech.v3.datatype and tech.v3.datatype.functional namespaces. Similarly to tmd, Cursive and Calva users now get full intellisense help with these main namespaces. Furthermore the FFI bindings now support linting with clj-kondo.

One thing about the readme that is incorrect - [dtype-next](https://github.com/cnuernber/dtype-next)'s ffi does in fact support callbacks :-). It is used as the backend to [libpython-clj](https://github.com/clj-python/libpython-clj) where you certainly can call clojure functions from python.

There are some interesting efforts concurrent with scicloj work by Chris Nuernberger specifically dtype-next, and the earlier tech-jna stuff. It's the same stuff underlying libpython-clj and libjulia-clj. recent talk.

This is great work. One of the things that has been on my mind working through our numerics stack is how to extend the number tower to complex numbers or more generally to arbitrary algebras. This project seems to me to be sort of a type-system-in-a-box that we can use to add arbitrary typing to Clojure where necessary/ideal. Thanks for sharing.