Loading…
4
Back in 2019 I had to pick between TensorFlow and PyTorch for a cancer screening model at a lab in Austin
I went with TensorFlow because our whole team already knew it and the deployment tools felt safer for hospital servers. Three months later we hit a wall with custom layers and I kept eyeing PyTorch tutorials at 1am wishing we had switched sooner. Did anyone else get locked into a framework early and just ride it out instead of starting over?
1 comments
Log in to join the discussion
Log In1 Comment
miam1124d ago
The part that gets me is the switching cost math people never do honestly. Everyone talks about framework features but the real wall is your validation data pipeline and the FDA paperwork that references specific model outputs. Rebuilding that stuff in a new framework is not a weekend project, it is a six month project you have to explain to hospital lawyers. So you end up staying because the code works and the audit trail exists, not because the framework is better. The tutorials at 1am are just grief for the road not taken, and that grief never really goes away.
6