The ML space is evolving at a crazy pace, and I totally get that people who are deep in the field with real deadlines don’t have the bandwidth to adopt yet another framework. That’s why I’m not aiming to build a framework from scratch to compete, but rather to build a productivity and automation layer on top of existing tools like Metaflow.
Cargo culting netflix’s tech is usually a bit questionable but maybe not this time!
> The ML space is evolving at a crazy pace, and I totally get that people who are deep in the field with real deadlines don’t have the bandwidth to adopt yet another framework. That’s why I’m not aiming to build a framework from scratch to compete, but rather to build a productivity and automation layer on top of existing tools like Metaflow.
Sorry looks like this was my Lack of proper reading!
More concrete feedback:
- Docs link in readme 404s
- get to code/demonstration or your unique advantage as fast as possible in README (“complete pipeline code” on this page: https://brkcvlk.github.io/MLFCrafter/getting-started/first-p.... ?)
- if its built on metaflow be upfront about it? The value prop would be saving time/enforcing a golden path?
- “<x>Crafter” is probably noise that should be deleted
- can you collapse some of the setup lines
- it might be nice if you could feed inputs to the pipeline as a whole, then you could use the context management APIs in Python (“with” IIRC?)
[EDIT] yeah metaflow has definitely been around for a while!