I've looked at PAQ8L source code here https://github.com/JohannesBuchner/p...aq8l/paq8l.cpp and saw that SSE/ APM (I guess APM is PAQ8's name for SSE) works on linear probabilities. That will (I think) give big rounding error when dealing with skewed probabilities (like 0.01 or 0.99). Shouldn't SSE work better on stretched probabilities?
Results 1 to 3 of 3
-
21st April 2018, 01:49 #1Member
- Join Date
- Jun 2009
- Location
- Kraków, Poland
- Posts
- 1,505
- Thanks
- 26
- Thanked 136 Times in 104 Posts
Why SSE in PAQ8 works on un-stretched probabilities?
-
21st April 2018, 02:19 #2Member
- Join Date
- Sep 2015
- Location
- Italy
- Posts
- 290
- Thanks
- 120
- Thanked 168 Times in 124 Posts
Doesn't the line 1259 pr=stretch(pr); stretch the probability?
-
Thanks:
Piotr Tarsa (21st April 2018)
-
21st April 2018, 02:31 #3Member
- Join Date
- Jun 2009
- Location
- Kraków, Poland
- Posts
- 1,505
- Thanks
- 26
- Thanked 136 Times in 104 Posts
Yes
Somehow I missed that. Perhaps the repeating stretching and squashing misled me.
Reply With QuoteSimilar Threads
-
How WinZip/WinRar Works ?
By imransuet in forum Data CompressionReplies: 3Last Post: 4th August 2017, 07:01 -
SSE, BMI do not accelerate LZ77 (un)compression
By lz77 in forum Data CompressionReplies: 12Last Post: 23rd June 2016, 11:24 -
(Reversed) entropy coding where decoder doesn't have to know the probabilities
By Jarek in forum Data CompressionReplies: 1Last Post: 17th February 2014, 23:05 -
SSE(o2,o4) CM coder
By Shelwien in forum Data CompressionReplies: 0Last Post: 4th February 2011, 13:50 -
Mixer vs 2D SSE
By Shelwien in forum Forum ArchiveReplies: 26Last Post: 8th March 2008, 13:58
