This week was a bit of a slow one; playing with some more compression stuff.. but in the weekend I ended up finally going through the last bits of the x86 byte categorizer for the 64k packer, and finally got fixed all of the remaining visible bugs!
This isn't that useful on its own, but when hooked up to a model for a compressor, should be able to provide some more useful information in order to predict what comes next in the bitstream. Particularly, in my compressor, the idea is to use the current byte category to select different contexts/weights in the model, different APM stages, etc.
Unfortunately, when hooking it all up, while I was able to see some gain, I wasn't able to see that much. Even after trying to implement some basic indirect models, which ended up hurting compression..
So not that much after all, but some kind of progress at least.. :)