This was running far more slowly than I expected locally, so I was looking at more logging and memory usage. (Turns out a rogue conda instance was bogging the machine down).
Worth noting that spike-in synthetic control detection is a bit slower than I'd like when using no or a very low abundance filter, but we're talking something like a minute overhead.
It seems d00a3ae06b8820e158a7cf5b87056747919c31d8 does actually help with memory usage when trying no or a very low abundance filter (testing Python 3.10 on macOS).
This was running far more slowly than I expected locally, so I was looking at more logging and memory usage. (Turns out a rogue conda instance was bogging the machine down).
Worth noting that spike-in synthetic control detection is a bit slower than I'd like when using no or a very low abundance filter, but we're talking something like a minute overhead.