PHP has decent FFI, nothing is stopping you from using the same libraries as you would with Python. Here's someone's quick hack as an example: https://github.com/dstogov/php-tensorflow
For an interpreted language PHP itself is ridiculously fast and the VM is rather small so you can use something else coughElixircough for parallellisation. I use it all the time for data wrangling stuff and database imports because it's robust, fast and PsySh is a pretty neat environment.
The array data structure is quite nice too. It's built on simple parts that are foundational to the VM itself, and very flexible, similar to lists in Lisp-like languages but without the seek lag when data grows due to the indexing.
It's very popular, as in a lot of businesses use it, it's just not fashionable.
I think it's a great tool to have. It had gradual typing before it was cool. You can type in like a page of code including the layout and render whatever in a PDO-supported database on a web page, served by the builtin web server, which is great for data exploration and things like SQL optimisation. At the moment I'm handling some data flows and conversions in a project with something like a terabyte of email and office documents that need to go into RDBMS, because there are some liberally licensed lightweight libraries (in contrast to the bulky stuff in Jakarta-land) and the performance is good enough to not be a bottleneck.
Edit: And when a library isn't good enough, I can usually trivially fix or extend it because it's in a familiar language and written by a simple minded person like myself.
There's a degree of clunkiness and incoherence in built-in API:s that might be off-putting at first but the included batteries and PsySH make for a quite decent tool anyway.
There were many many times I'd start writing a bash script, but then switch it to a PHP script. I've done this so many times that now I just start writing in PHP.
For an interpreted language PHP itself is ridiculously fast and the VM is rather small so you can use something else coughElixircough for parallellisation. I use it all the time for data wrangling stuff and database imports because it's robust, fast and PsySh is a pretty neat environment.
The array data structure is quite nice too. It's built on simple parts that are foundational to the VM itself, and very flexible, similar to lists in Lisp-like languages but without the seek lag when data grows due to the indexing.