| Thu, 15 Jan 2015 17:29:49 +0000 |
Chris Cannam |
Attempt naive silence penalty -- not currently working, but the plots are food for thought
silence_penalty
|
| Thu, 08 Jan 2015 12:11:27 +0000 |
Chris Cannam |
Separate out the raw & conditioned feature outputs (previously only conditioned was available, but we want raw for our tests). Plus some optional debug output
refactors
|
| Thu, 18 Dec 2014 17:56:54 +0000 |
Chris Cannam |
The finder is supposed to use normalised path-cost when calculation expand direction (as in Java implementation). Also, provide a way to query the forward path.
refactors
|
| Thu, 04 Dec 2014 13:23:10 +0000 |
Chris Cannam |
Feature size is not actually needed in Matcher
feature_conditioner
|
| Thu, 27 Nov 2014 16:50:14 +0000 |
Chris Cannam |
Tidy
refactors
|
| Thu, 27 Nov 2014 16:05:19 +0000 |
Chris Cannam |
Ah, it's subtler than that -- we have to handle the case where this is not the first matcher as well, if we're to get precisely the same results
refactors
|
| Thu, 27 Nov 2014 15:56:12 +0000 |
Chris Cannam |
Fix inconsistency (with Finder) in choice of preference where two costs come out equal
refactors
|
| Thu, 27 Nov 2014 13:02:28 +0000 |
Chris Cannam |
Remove some debug out
refactors
|
| Thu, 27 Nov 2014 12:59:23 +0000 |
Chris Cannam |
Fix diagonal handling: there was an implicit fixed diagonal weight of 2 in comparison of diagmin+distance to othermin directly
refactors
|
| Thu, 27 Nov 2014 12:54:20 +0000 |
Chris Cannam |
Move some functions about
refactors
|
| Thu, 27 Nov 2014 12:08:16 +0000 |
Chris Cannam |
Previous commit was a mistake: the ahead-of-time business is in Finder::getExpandDirection. In fact we were failing to swap advance directions in forward path when writing to the "other" finder. This does not actually affect the backward path calculation, but it does mean we can restore the sanity check.
refactors
|
| Thu, 27 Nov 2014 10:53:00 +0000 |
Chris Cannam |
More error checking (check advance as well as cost)
refactors
|
| Thu, 27 Nov 2014 10:49:11 +0000 |
Chris Cannam |
Expose diagonal weight as a parameter
refactors
|
| Wed, 19 Nov 2014 12:13:28 +0000 |
Chris Cannam |
Merge from branch refactors
cheap_diagonals
|
| Wed, 19 Nov 2014 11:59:03 +0000 |
Chris Cannam |
Pull out feature extractor calls from Matcher, remove MatchFeeder, have only the feeder-from-features and use that in MatchVampPlugin
refactors
|
| Wed, 19 Nov 2014 10:18:19 +0000 |
Chris Cannam |
Pull up Matcher set/get to public API, use only public API in Finder
refactors
|
| Wed, 19 Nov 2014 09:17:58 +0000 |
Chris Cannam |
Encapsulate get/set, add range and init checks
refactors
|
| Tue, 18 Nov 2014 16:35:15 +0000 |
Chris Cannam |
distYSize no longer needed, and some tidying and checks
refactors
|
| Fri, 14 Nov 2014 12:35:36 +0000 |
Chris Cannam |
Weight of sqrt(2) in both cases
cheap_diagonals
|
| Fri, 14 Nov 2014 10:22:27 +0000 |
Chris Cannam |
Do not penalise diagonals
cheap_diagonals
|
| Fri, 14 Nov 2014 10:09:10 +0000 |
Chris Cannam |
If the distance metric is a float, the total path cost should be a double (otherwise discriminatory precision is lost as the overall total accumulates)
refactors
|
| Fri, 14 Nov 2014 09:45:56 +0000 |
Chris Cannam |
Fix narrowing float conversions
refactors
|
| Thu, 13 Nov 2014 15:09:04 +0000 |
Chris Cannam |
Some fixes: int -> float
refactors
|
| Thu, 13 Nov 2014 14:43:03 +0000 |
Chris Cannam |
Replace char dist (inc dir in masked section) with float dist + separate dir
refactors
|
| Thu, 13 Nov 2014 13:53:52 +0000 |
Chris Cannam |
refactor: m_ prefix
refactors
|
| Thu, 13 Nov 2014 13:19:53 +0000 |
Chris Cannam |
Fix to resizing logic. Regression test passes.
refactors
|
| Thu, 13 Nov 2014 12:57:04 +0000 |
Chris Cannam |
Merge from default branch
refactors
|
| Thu, 13 Nov 2014 12:55:29 +0000 |
Chris Cannam |
Avoid freqMap overruns for unusual sample rates
|
| Thu, 13 Nov 2014 12:50:54 +0000 |
Chris Cannam |
Make use of FeatureExtractor in Matcher
refactors
|
| Thu, 13 Nov 2014 09:50:14 +0000 |
Chris Cannam |
Move source to src
refactors
base
Matcher.cpp@cdeca0b49e1d
|
