Tony
Yesterday I tackled credits and aggregate credits. Today I did something new to focus on weights.
Instead of spending 2 to 3 hours going back and forth with AI, I decided to set up an engineering agent named Tony. I instructed Tony to adjust the weights based on the work we have done so far.
Before today, every shared person counted the same. Walter White and a one-episode guest were both worth 1. Now each person counts as the share of the show they made. Walter White is in all 62 episodes of Breaking Bad, so he counts as 1.0. A one-episode guest counts as 1 of 62. This makes more sense to me, and it equalizes long-running series and mini-series. The cast and crew also rank on one list now.
The recommender so far
- Day 1: connect shows that share cast members and count the shared people.
- Day 2: add crew to the graph and stop counting casting offices.
- Day 3: pull every episode credit from TMDb. The data got 176 times bigger and the recommender got worse.
- Day 4: weight every shared person by the share of the show they made, and merge cast and crew into one ranked list.
Observation
I also ran my gstack QA/QC skill today and it found two bugs. One bug let a person count for more than a whole show. TMDb credits Mariska Hargitay with 595 episodes of a 594-episode series. I cannot recommend the gstack skills enough. They have helped me discover things I otherwise would not think about, or that AI would not catch on a first or second pass.
Next Up
Tomorrow I am going to follow up on some of the ingest work. I wanted to work on the detail page, but I am traveling this weekend and want to give the detail page a day’s worth of attention.
See you tomorrow for Day 5!
For a more in-depth explanation, read this ADR.