DEV Community

Discussion on: Writing machine learning code on 20 year old software on 30 year old hardware

Collapse
 
sqlrob profile image
Robert Myers

I was running OS/2 on a 386SX, it was a DOS neural network program that I don't remember the name of at all, it was hardcoded for backprop networks.

I was attempting to see if I could identify coding vs. non-coding regions of DNA. Unfortunately, it was no better than a coin flip when all was said and done.

Thread Thread
 
goyder profile image
goyder

Yowza - that's an unfortunate finding for a training run that lasted weeks. That'll keep me a bit more humble when I get grumpy or impatient about deep learning runs that take an hour or two.

Doing bioinformatics in the early 90s must be have been a pretty interesting experience! Was this kind of analytics/computation role your core role or just a side interest? What was the data and computational power availability?

Thread Thread
 
sqlrob profile image
Robert Myers

That was for the term project in a grad level Neural Network class, not a job. The computational power was my own PC, so rather limited :D. The data was the sequences from a few genes tagged as intron or exon, as even back then you could download sequences from GenBank.