Results 1 to 4 of 4

Thread: State machines

  1. #1
    The Founder encode's Avatar
    Join Date
    May 2006
    Location
    Moscow, Russia
    Posts
    3,982
    Thanks
    377
    Thanked 351 Times in 139 Posts

    Question State machines

    It's time to make clear to everyone, how to use state machines instead of counters.

    Considering CCM, we can use something like that:

    pr=table[pr][bit];

    i.e. update a probability using pre-generated table. A question, how to generate, and the most importantly, how to optimize that table?

  2. #2
    Administrator Shelwien's Avatar
    Join Date
    May 2008
    Location
    Kharkov, Ukraine
    Posts
    3,366
    Thanks
    212
    Thanked 1,018 Times in 540 Posts
    Gribok just answered that in qlfc thread - that approach with contexts works for counters too.
    Counters in paq and ccm are ad hoc though, the only public program with tunable fsm is m1 afaik.

  3. #3
    The Founder encode's Avatar
    Join Date
    May 2006
    Location
    Moscow, Russia
    Posts
    3,982
    Thanks
    377
    Thanked 351 Times in 139 Posts
    I'm talking about that tunable FSM!

  4. #4
    Administrator Shelwien's Avatar
    Join Date
    May 2008
    Location
    Kharkov, Ukraine
    Posts
    3,366
    Thanks
    212
    Thanked 1,018 Times in 540 Posts
    I meant this:
    Quote Originally Posted by Gribok
    I use following idea:
    1. Collect statistics like [Pos, Bit] for all contexts.
    2. Use greedy approach for the next step.
    3. In each step I try to take two contexts and merge them in single one. During merge I sort statistics by Pos and then use adaptive arithmetic for Bits to measure compression.
    4. Stop merging if number of contexts <= 256
    And as to m1, its for toffer to explain, I don't really understand what he does there.
    But afaiu its also based on n0,n1 counts (like paq and ccm), so I think that bruteforce
    quantization like quoted above would be better.

    Anyway, counter states can be considered contexts for static predictions, so its
    the same task as context quantization.

Similar Threads

  1. Sac: (State-of-the-Art) Lossless Audio Compression
    By Sebastian in forum Data Compression
    Replies: 55
    Last Post: 12th November 2019, 12:24
  2. Compression state-of-art discussion with C.Bloom
    By Shelwien in forum Data Compression
    Replies: 7
    Last Post: 27th October 2011, 07:18
  3. Virtual test-machines
    By Vacon in forum Data Compression
    Replies: 7
    Last Post: 15th April 2009, 00:19
  4. Lines are blurring between humans and machines
    By encode in forum Forum Archive
    Replies: 0
    Last Post: 27th May 2007, 21:20

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •