Closed dzeber closed 5 years ago
@birdsarah Thanks for taking the time to review this in depth - I appreciate the detailed comments, and I think they've really helped me hone down the focus.
I've made the requested changes to GuidGiudRecommender.md
--updates to the other doc will follow.
I've also added a few details to the README about the taar_lite_guidguid.py
processing. One thing I wanted to call out is that I didn't see any pruning of rare combinations in there (I think we'll be doing this as a recommender treatment instead), so I removed that bullet point.
@birdsarah @mlopatka @crankycoder Thanks everyone for the detailed review! I believe I've addressed every comment, either through changes or by replying.
In response to Sarah's overall requests:
- obsessively consistent use of terms, and ideally a definitions section. This is to help people like me who are new and really don't know whether words are interchangeable and so it reduces the cognitive load.
AFAICT I've used terminology consistently throughout, and relevant terms are now all defined (and usually bolded). Definitions are often inline in the text as opposed to in a formal glossary.
- Mini examples for each of the treatments for people like me who are more concrete / visual thinkers
I've added an example for each treatment, some with graph diagrams. ASCII art FTW! I'd love to find a better way to create diagrams, but ideally it would render nicely in Github markdown without relying on external tools.
- Do not mix-in production / deployment details. They can be written up in a seperate doc about app/production.
Done - these have been removed and parked for future use.
Adding some doc on the logic underlying the GUID-GUID recommender, treatments, graph representation etc.
I've created two doc pages:
GuidGuidRecommender.md
: description of theGuidGuidCoinstallRecommender
at a fairly high level, making the connection to implementationGraphRepresentation.md
: more technical details about the graph representation of the coinstallation data and recommendation algorithm. I've added detailed descriptions of each of our implemented treatments here as well. This is more for our own reference as we start thinking about graph-based evaluation metrics.There is also some minor cleanup of the main README.
r? @mlopatka @birdsarah