LZIFY
Arguments: text niter &optional type [generic-function]
Builds a pattern dictionary containing a statistical model.
inputs:
- a function selecting the analysis information inside a symbol.
- a function selecting the synthesis information inside a symbol.
- a function that reconstructs the symbol from analysis and synthesis information (will be used by LZGenerate).
output:
a LZ continuation tree
a LZ pattern tree
a LZ pattern tree
LZify takes a list of anything considered as an ordered sequence.
It then builds a pattern dictionary that encodes patterns of various lengths
discovered over this sequence, as well as the conditional probabilities
that a certain pattern be followed by certain elements
If niter is greater than 1, the analysis of the sequence is iterated <niter> times,
each time skipping the next element on the left of the sequence.
niter > 1 increases the number of patterns discovered. It is equivalent to analyzing
a longer sequence, thus increasing the statistical properties (redundancy).
Empirical experience shows that niter = 4 is good value for such data as bach-like counterpoint
or jazz chorus.
Then LZify builds the continuation tree. This tree is another representation of the pattern
dictionary, suited to generation features : the branch are reversed, in order to ease the search
for the maximum context, and the continuations of each context are explicitly represented, linked
with their corresponding probabilities.
It then builds a pattern dictionary that encodes patterns of various lengths
discovered over this sequence, as well as the conditional probabilities
that a certain pattern be followed by certain elements
If niter is greater than 1, the analysis of the sequence is iterated <niter> times,
each time skipping the next element on the left of the sequence.
niter > 1 increases the number of patterns discovered. It is equivalent to analyzing
a longer sequence, thus increasing the statistical properties (redundancy).
Empirical experience shows that niter = 4 is good value for such data as bach-like counterpoint
or jazz chorus.
Then LZify builds the continuation tree. This tree is another representation of the pattern
dictionary, suited to generation features : the branch are reversed, in order to ease the search
for the maximum context, and the continuations of each context are explicitly represented, linked
with their corresponding probabilities.
