People

Dan Ashlock

New Code and Data

• The genetic algorithm, using a hybridized initial population, GA_2003_TwoClass.cpp
• Training data: primer2000.dat.
• First run classifiers: BestClA.dat
• The header vaut.h and the code vaut.cpp for the finite state machine class.

g++ -lm -O3 -o evolve.exe GA_2003_TwoClass.cpp vaut.cpp

Place evolve.exe, primer2000.dat, BestClA.dat in a directory together, execute evolve.exe, and send me the resulting file classAut.dat.

Code

To run the machine learning algorithm you will need one of the two mail source files below, the vaut class of finite state machines, and a traning data set (of which we have one available at present).

• The genetic algorithm, using a random initial population, PrimClass.cpp
• The hybridizing genetic algorithm, reading in a file a good predictors, PrimClass2.cpp
• The header vaut.h and the code vaut.cpp for the finite state machine class.
• The training data Primerset.txt in the format "N ", one per line. The number N codes 0=good primer, 1=multiple band primer, 2=failed to amplify.
• A new larger training data set Primerset2.dat in the same format.
• A full 10k data set Primers10K.dat in the same format.

---

Using the code

g++ -lm -O3 -o evolve.exe PrimClass.cpp vaut.cpp

After running the genetic algorithm several files will be present. There include run#.dat files which contain the mean, deviation of, and maximum fitness for each generation of each run as well as a classAut.dat file that contains the bes automata found in each run together with its fitness.

To perform hybridization, use the PrimClass2.cpp source file and include a file BestClA.dat in the directory. This file is in exactly the format of the classAut.dat file produced by PrimClass.cpp. Simply copying the file is the simplest way to procede.

Best-of-Run Data:FSMs

• Baseline with the original code.
• has -1s in the left column of its weight matrix.
• has -2s in the left column of its weight matrix.
• is the first run of the hybridized fitness function code using the absolute value of the original payoff matrix.
• uses the second payoff matrix that I pulled from nowhere.
• The best of run for the ramped fitness function experiment.
• The best of run for the basic experiment with 32 states.
• The best of run for the cross-validation experiment.

Pictures(Postscipt version in PS link)

PS
PS
PS
PS
PS
PS
PS