HackerTrans
TopNewTrendsCommentsPastAskShowJobs

alxmrs

64 karmajoined 5 lat temu
https://alex.merose.com

Submissions

Show HN: I implemented a neural network in SQL

github.com
87 points·by alxmrs·12 godzin temu·17 comments

comments

alxmrs
·8 godzin temu·discuss
I need to better understand your first question before I can comment. In theory, we could work with MADlib too -- what we do is port scientific data, which typically is a "tensor" or Nd array, into a tabular view. I believe you know relational theory better than I do, I am still fairly new to the field.

WRT ISO/IEC 9075-15:2023: This is the standard established from rasdaman, IIUC. I reject this approach (which treats arrays as a column type), and instead adopt one inspired by Michael Stonebraker's SciDB (which treats arrays as tables themselves). For an in depth review of the topic, I recommend this NSF paper: https://par.nsf.gov/servlets/purl/10545560
alxmrs
·10 godzin temu·discuss
Totally. I can’t wait to take this to https://hytradboi.com
alxmrs
·10 godzin temu·discuss
https://bsky.app/profile/al.merose.com/post/3mpz4njpcvk2o
alxmrs
·11 godzin temu·discuss
I would have had the same gut reaction as you lol.

You’re spot on. I think that SQL, as a data oriented and logic PL, might be ideal for writing tensor programs.