peter-ebert

I've added basic bit packing to the frequencies table, so long as the dictionary doesn't dominate like in a pangram it's slightly smaller than two arithmetic (static and adaptive) examples I could find online. I understand this isn't the best that arithmetic can do, next I'll look into that.

Never was trying to compare to NNs or PPM (or even simple LZ), since all this algorithm does is look at the frequency counts.

Agree it's not very practical and very expensive calculations, just fun to work on.

Thanks I'll work on that. One thing though > The linked example can encode _any sequence_, not just ones that have an equal number of ones and zeros.

The compression I made can also encode any sequence of bytes, see the testinput folder (caveats in the FAQ are because of the implementation specifics not the algorithm), it just also has to store the freq table which I'm working on compressing too. It has no requirement for equal number of 1s and 0s, that was just an example in the description that I guess did more bad than good.

*I mean input2, sorry long day.

I appreciate the input. I did not mean to imply I could encode a random stream of symbols/characters, that is absolutely valid. I was approaching this as a compression technique for something like a text file, where the symbol counts are known at compression time.

Thanks I'll correct, I didn't know you could adjust the symbols as you go with arithmetic coding, got an example of that?

I was using what I seemed to be the standard with other encoders, storing the ratios is roughly the same cost compared to other compressors, a few extra bits for exact counts. But I do see how this isn't i.i.d. so I'm fixing that.

Got it, I was thinking of the locations as being i.i.d. in that you do not have any information on their location. You and others are saying that adjusting the symbol counts as you go would give the same performance to other algorithms. Looking into that, thanks.

thanks for the constructive feedback!

Whenever you're compressing a file on your computer the frequencies are known beforehand. Agree this does not apply to noisy-channel coding, only the Shannon source coding theorem in lossless compression. Good point though I'm not sure if this would be considered an entropy encoder, as previous values do have some impact. Any more info I should look at for that?

Yeah this was a big fear in posting. No I'm not experienced in academia or this would be a paper, I code things for fun. Have you tried looking at the math for multinomials vs Shannon? Or running the code? Please do point out my mistakes in the math.