Re: combining probabilities from different models

moogie <budgetanime@xxxxxxxxxxxxxx> writes:
Hi,

Most of my forays into compression have been based on using one
particular model to generate a probability for a given symbol. I now
have multiple models that each give a proability for a given symbol.

My question is how does one combine these probabilities to form one
proability?

Is it as simple as using the mean of the probabilities?

I defer to JDA, as he is obviously closer to the field than I,
but one thing that comes to mind is to have a monitor which
evaluates the probability of each of the models being the better
model, and to take a weighted average depending on what the
monitor is saying. For example, if you're compressing executables,
one model might be a good predictor for the machine code generated
by a high level language, and another model might be a good
predictor for the kind of static data the program has. Different
models will be better at different times, and the monitor should
reinforce the influence of the model which is better.

James - is your paradoxical data related to Simpson's Paradox?

Phil
--
Dear aunt, let's set so double the killer delete select all.
-- Microsoft voice recognition live demonstration
.

Relevant Pages

  • Re: Paying for assistance
    ... forget about the input being 8 bit sequences. ... This is just uninteresting, and forget about the dictionary, this is only a technicality and we can assume without loss of generality that the input is sorted by probability. ... The expected lengths of files 1,2,3 are then given by multiplying the expected bit count by the entropy of the given file, i.e. ... them 'bait' for a compression algorithm. ...
    (comp.compression)
  • Compression is Equivalent to General Intelligence
    ... intelligence and compression that needs to be cleared up. ... (encoded as strings). ... The only possible probability distribution over ... The problem of finding the shortest program consistent with an agent's ...
    (comp.ai)
  • Re: kosher jpgs
    ... compress a jpg with its proprietary jpg compression method. ... It is already compressed using a lossy compression algorithm. ... Lossy compression allows you to throw out entropy, while lossless only minimizes non-entropy data. ... definition of probability, and probability is only definable with ...
    (comp.lang.java.programmer)
  • Re: How much lossless compression is possible in images?
    ... This isn't really a compression algorithm in the sense being discussed ... A compression algorithm is simply a substitution algorithm. ... substitution of messages from A is forced by equal probability for all ...
    (sci.math)
  • Re: Compression by Use of Info String
    ... messages, you need to establish that probability of occurrence of each message, ... The PPMD encoding assumes that there ... So "Optimal Compression" is at best a theoretical idea, ... practical approach for achieving in practice. ...
    (comp.programming)
Loading