2 ms·
You'd probably want to use something like CABAC rather than Huffman, since it deals better with dynamic probabilities. https://en.wikipedia.org/wiki/Context-ada
by Retr0id 3y ago
You'd probably want to use something like CABAC rather than Huffman, since it deals better with dynamic probabilities. https://en.wikipedia.org/wiki/Context-adaptive_binary_arithmetic_coding https://en.wikipedia.org/wiki/Context-adaptive_binary_arithm...
- gus_massa 3y agoIt looks like a very interesting comparison. I still not sure how to tranform StockFish points to probabilities. (Perhaps proportional to exp(-difference/temperature), where temperature is a magical number or perhaps it's like something/elo??? There are too many parameters to tweak.)
- Retr0id 3y agoGiven OP's dataset, you could probably figure out how to map them, experimentally (e.g. by making a histogram)
- gus_massa 3y agoI was imagining downloading the lichess database and trying different strategies to compress it, It's too much work for me, but I'd love to read a blog post if somsome does that.