To be fair, we did try. This was initially (2016) spearheaded (and the domain is still paid for) by Theam , then a small but active community grew around it, I was very active as an editor/hype man for the community (including running a workshop at ICFP’17 and pulling people in at ZuriHac and Haskell Exchange), we started building things, but then it sort of fell apart, people disappeared and the project ground to a halt.
And this is not meant to take credit from the crazies visionaries who have built machine learning libraries in this language. DataHaskell was simply meant to organize this effort. I have theories on why didn’t succeed (yet?), certainly have my share of the blame for not coordinating communications more effectively.
But the truth is that this stuff is super f+++++g hard to start with, it’s a lot of unglamorous work and additionally when you put 5 Haskellers in a room you get 12 alternative APIs for doing even the simplest task. We are simply allergic to standardization.