[root]/src/dr/inference/markovchain
Author | Changes | Lines of Code | Lines per Change |
---|---|---|---|
Totals | 69 (100.0%) | 3349 (100.0%) | 48.5 |
sibon.li | 5 (7.2%) | 1223 (36.5%) | 244.6 |
Sebastian.Hoehna | 7 (10.1%) | 728 (21.7%) | 104.0 |
jheled | 15 (21.7%) | 481 (14.4%) | 32.0 |
alexei.drummond | 5 (7.2%) | 357 (10.7%) | 71.4 |
msuchard | 4 (5.8%) | 292 (8.7%) | 73.0 |
rambaut | 30 (43.5%) | 263 (7.9%) | 8.7 |
gerton.lunter | 1 (1.4%) | 4 (0.1%) | 4.0 |
Sibon.Li@gmail.com | 2 (2.9%) | 1 (0.0%) | 0.5 |
Changes to MarkovChainDelegate to allow it better access and privileges to MarkovChain
1 lines of code changed in 2 files:
If +Inf or NaN in MarkovChain, it now converts these to -Inf so it can continue if the user has chosen to ignore errors.
10 lines of code changed in 2 files:
Porting r5138 from 1.7 branch
16 lines of code changed in 1 file:
Changed compound likelihood to return the actual special values (i.e. +Inf, NaN or -Inf) up the chain. Added 'todo' comment in MarkovChain suggesting a check for +Inf and NaN. Need to review this idea.
5 lines of code changed in 1 file:
Added an interface for delegating tasks in the Markov Chain
72 lines of code changed in 2 files:
Added ability to co-opt an existing root height in StarTreeModel
9 lines of code changed in 1 file:
Finally fixed the diagnostics strings displayed after an incorrect restore so that they are genuinely before and after (as opposed to both after).
9 lines of code changed in 1 file:
Commented out a couple of debug printfs
4 lines of code changed in 1 file:
Removing wrongly located files.
0 lines of code changed in 1 file:
Reverting changes in 4754 - MarkovChain is not designed for subclassing.
19 lines of code changed in 1 file:
Commented out all of my last commit until issues resolved
536 lines of code changed in 1 file:
Added classes to enable implicit convergence assessment checks using BLUE
595 lines of code changed in 1 file:
Changes to encapsulation in MCMC.java and MarkovChain.java. MarkovChain.java no longer final and some private variables changed to protected
20 lines of code changed in 1 file:
Upgraded chain length to a 64 bit integer so that chain lengths > 2.1 billion can be done. This addresses issue 540. Runs OK, accepts long chain lengths but not actually tested a run of > 2.1 billion.
11 lines of code changed in 3 files:
Cleaned up full evaluation test error reporting.
5 lines of code changed in 1 file:
Started work on caching in ATL
2 lines of code changed in 1 file:
Fix for dynamic scaling with BEAGLE?
17 lines of code changed in 1 file:
Added some simple timing round the evaluation just after each operator which is stored by the operator and reported in the operator analysis table. On this note, I have added an attribute to <mcmc operatorAnalaysis="analysis.file"> which gives the name of the file to write the analysis to. If this isn't there then the file isn't written. We need to add this to BEAUti to use the same file name stem as the log and trees file (with an extension '.ops'?).
10 lines of code changed in 1 file:
Replaced 'preBurnin' with an autoOptimizeDelay (coercionDelay). The rationale for this is that the preBurnin meant that the initial state was not in the log files, it was unecessary and it cause a small inefficiency as the JIT compiler re-optimized after the preburnin chain had finished. We also didn't see warning messages until after the pre-burnin period. There is a new attribute to the MCMC parser - autoOptimizeDelay (to which the old preBurnin attribute now maps so old XML will run).
8 lines of code changed in 1 file:
generify and auto intelliJ
5 lines of code changed in 2 files:
casting problem
10 lines of code changed in 1 file:
Even slower at startup - have more detailed printouts
4 lines of code changed in 1 file:
BEAGLE GammaSiteModel now accepts a mutationRate for compatibility with some BEAST XML.
2 lines of code changed in 1 file:
First step in development of a generic version of Parameter, called Variable for now. Parameter now implements Variable<Double> but eventually all dr.inference.* will be based on Variable rather than Parameter. At least that is the idea anyway...
5 lines of code changed in 1 file:
refactoring to remove illegal dependencies (dr.inference should not depend on dr.evo*.*) and tidying
37 lines of code changed in 1 file:
comment out operator experimental code
60 lines of code changed in 1 file:
Made the threshold at which the full evaluation test fail into a static constant at the top of the class.
4 lines of code changed in 1 file:
Control number of operator evaluation checks via minOpsFullEvaluations.
Add 'minOpsFullEvaluations="0" fullEvaluation="0"' to the <mcmc> attributes to disable checks altogether.
6 lines of code changed in 1 file:
Temporary solution before full code review: if fullEvaluationCount == 0, then switch off usingFullEvaluation
2 lines of code changed in 1 file:
Check evaluation after accept move as well. chaeck until at least N (=1) accpets and rejects for each operator.
22 lines of code changed in 1 file:
Slight modification of when the fullEvaluation test is performed.
1 lines of code changed in 1 file:
Tidying up to resolve more the confusion between the Gibbs and Metropolized Gibbs Proposals.
2 lines of code changed in 1 file:
Change name of SimpleGibbsOperator to SimpleMetropolizedGibbsOperator to avoid confusion.
4 lines of code changed in 1 file:
Full evaluation continues until every operator has done at least one operation. If an error has occurred in this time, then a runtime exception is thrown.
32 lines of code changed in 1 file:
Cleaned up MarkovChain a little more. There were some debug printfs surrounded by if statements. Made the flag that turned these on into a static final - this should mean the if statements will not be compiled in if false.
Added comment about odd coding surrounding Gibbs operators - don't want to fiddle but looks suspect to me.
16 lines of code changed in 1 file:
Added an additional test of full evaluation of the likelihood after an operator. This should detect operators that don't fully flag up what to update.
I also removed a likelihood.makeDirty() call that was removing efficiency savings on restore. This was presumably a work around for a bug in another class. We should expect the test of full evaluation after restore test to fail on some models (but who knows what? This call was masking the test of correct restore). Benchmark.xml works with this call removed.
25 lines of code changed in 1 file:
cosmetic
266 lines of code changed in 1 file:
Added some stuff to get the MGS working.
288 lines of code changed in 1 file:
Fixed problem generating XML to sample from the prior in BEAUti.
8 lines of code changed in 1 file:
automatic code improvments (intellij)
1 lines of code changed in 1 file:
(21 more)