EXP
EXP
EXP - surely, you must be joking
EXP - from 50% to 80% probability of not have a meta-gene right before an exon
EXP - right now the probability of no-meta / 0-meta / 1-meta was the same. Experimenting a bit with modifying it
Still cleaning up jade.env.PatternTracking
Getting jade.env.PatternTracking ready for experiments
EXP - comparing DGA to FGA / FGA-named / SGA as part of baseline for cyclicpatterntracking
Implementing DGA
Getting the DGA package started as a copy of FGA
EXP - tuned a bit the parameters of the baseline experiments on cyclic pattern tracking with FGA, FGA-naked, and SGA so that the latter would be able to find the first optimum. Results show that SGA beats both versions of FGA to the first optimum, but is then unable to keep tracking the new optima. As expected, FGA beat both SGA and FGA-naked. Reproduction of previous results successful.
REFACTORING - This code was no longer used since we now handle all parameters setting in the CfgParser via meta-programming
REFACTORING - just cleaning up things a bit
EXP - ICTAI97 paper suggests values for mirroring and folding that are way below what we used here; i.e. 0.001 vs. 0.05 and 0.01. We tried these values and it lead to FGA baseline performance to finally beat the SGA. The FGA now also converges slightly faster than the FGA-naked version that did not use mirroring or folding. Et voila! :)
EXP - uncommenting. The previous experiments showed that setting the rates to 0 indeed works
EXP - trying to comment out both folding and mirroring
experimenting - deactivating folding the hard way just to make sure we eliminate any side effects when comparing
refactoring - keep separate the bogus gene expression just in case we need it later for debugging purposes
Still not working; the FGA without any specific operators slightly outperform the SGA after the first epoch but when turning on mirroring and folding the FGA performance is crappy. Anyhow, here are the cfg files used for these experiments.
Still working on fixing the folding operator
Observed big differences between previous baseline results and this new version; tracked it down to bad implementations of the folding operator and, more specifically, its neutrality
Forgot to put back the mirroring and folding in evolve
ktSelect replaces recruitment and selection
Replacing selection and recruitment by ktSelect
Starting the merging of selection and recruitment in jade.ga.Algo
BUG FIX
BUG FIX sneaky bug preventing the Statistic objects to be given proper references to cfg and pop fields from Statistics
updating cfg
Implementing FGA
The cfg file that goes with the previous commit
Removing default class-level fields' initializations in jade.ga.Config
Default for seed value is 0 and should trigger a time-based randomization
readme.txt file for this branch never got committed
Fields marked as @ExperimentParameter should not be initialized outside constructors; this led to bugs fixed in previous commits. They should not either be initialized in the constructor since super() now ends up triggering the load() in CfgParser. Trying to initialize them after this does not result in a "default" value assigned in the absence of values in the cfg file, it instead results in overwriting any configuration via the cfg file by these initialization values. Another bug that was h...
updated the cfg file so we may test jade.fga. WARNING make sure to always use jade.fga.Config when using jade.fga.Algo
Fixing lil problem with the reification process in Run.
Placeholder implementation of FGA
updated the cfg for the test experiment so that it actually runs for 1000 generations. Tested by comparing bst.exp to results.sum; the jade stats and the ones resulting from dumping ECJSimpleStats and then analyzing them with the exp-analyze.sh script are the same. woohoo.
well that was stupid...
ECJSimpleStats implemented to produce the results.all directly
oops taking that back; jade.stats.ECJSimpleStats will provide us with the results.all directly
Configuring exp-analyze.sh to expect the output of jade.stats.ECJSimpleStats to be a bunch of job-XXX.out.stat.bstonly files
Adding jade.stats.ECJSimpleStats as a copy of BstStat to start off with
adding scripts used with ECJ and adapting them. Not yet tested since we have not added ECJShortStatistics in Jade.stats
Spring refactoring in preparation for FGA vs. McEA experiments
Removing debugging System.out.println
Integrating branch ITSP-2014-Davis improvements...
Moving Paul Burton ITSP work from the trunk to ...
US701 Broker verify garbage collection
US507.2 Broker Load Genotypes
US507.2 Load Genotypes Experiment
US507.1 Broker Save Genotypes
US507.1 Save Genotypes Experiment
US506.2 Broker Load Student Data
US506.2 Load Student Data Experiment
US506.1 Broker Saving Student Data
US506.1 Saving Student Data Experiment
US504 Broker Display Summary Data
US504 Display Summary Data Experiment
US503 Broker Genotype Records
US503 Genotype Records Experiment
US502 PSI Keep Student records
US502 Broker Keep Student Records
US502 Keep Student Records Experiment
US501 PSI Identify Genotypes
US501 Broker Identify Genotypes
US501 Identify Genotypes Experiment
US500.2 PSI Identify Students
US500.2 Broker Identify Students
US500.2 Identify Students Experiment
US500.1 PSI Get Student ID
US500.1 Broker Get Student ID
US500.1 Get Student ID Experiment
US405 PSI Feedback to Users
US405 Feedback to Users Experiment
US404 PSI Applying Transforms
US404 Broker Applying Transforms
US404 Applying Transforms Experiment
US403 Broker Model Transforms and answers key
US403 Broker Model Transforms
US403 Model Transforms Experiment folder
US401 PSI Checking for Correct Order
Broker consolidating common files in the broker
US401 Checking for Correct Order Experiment
US400 PSI Show Puzzle to user
US400 Broker Show Puzzle to user
US400 Show Puzzle to user
US311 PSI Puzzle in PSI
US311 Broker Puzzle in PSI
US311 Puzzle in PSI Experiment
US306.2 Broker reduced evaluation number for te...
US306.2 Add Author Experiment
Broker Fix Multiples
Fix Multiples Experiment
US310 Broker EA initialized from Broker
US310 Jade EA Initialzed from Broker
US310 EA initialize from Broker Experiment
US309 PSI Real ParsonsPuzzle
US309 Broker Real ParsonsPuzzle