3 ms·
> The coolest part IMO is how easy it is to plug in custom models for symbol probabilities. Usually a simple counter for each symbol is enough, but you can go c
by account42 4y ago
> The coolest part IMO is how easy it is to plug in custom models for symbol probabilities. Usually a simple counter for each symbol is enough, but you can go crazy and start predicting the next symbol based on previous ones.
PPM-based [0] compression can work quite well for text but on its own it's not enough to unseat the Lempel-Ziv family as the top general purpose compressor.
[0] https://en.wikipedia.org/wiki/Prediction_by_partial_matching https://en.wikipedia.org/wiki/Prediction_by_partial_matching
- PhilipRoman 4y agoYeah arithmetic coding on its own is not enough, typically you would have it as the last step, with dictionary and other compressors before it.