Darren Lewis

CS426/BMI226/EE392K

Using the GENESIS 5.0 code:

1. Create a directory in your Leland user space called CS426 or whatever you desire:

   mkdir CS426

2. Change to this directory:

   cd CS426

3. Run the following command to copy the source code to your user space:

   cp –rf /afs/ir/class/cs426/genesis .

   This will create a directory in your CS426 directory called genesis.
4. Change to this new directory:

   cd genesis

5. Compile and build the code using the following command (Note: not just make):

   make install

6. Create a file containing your evaluation function. Make sure your function matches the following prototype:

   double eval(char *str, int length, double vect[], int genes);
   
       where the different parameters are defined as the following:
   
   char *str; /* string representation */
   int length; /* length of bit string */
   double vect[]; /* floating point representation */
   int genes; /* number of elements in vect */
   

7. Run setup and fill in the appropriate information.
   • Use only options a, c, e, f, and g, and perhaps M. If you are running in binary string mode, do not use options f and g, because these deal only with floating point representation.
   • Leave the "scaling window", "generation gap", "dump" parameters, and "consecutive generations" at their default values (just press enter).
   For more information about the available options, please see genesis.txt. Feel free to experiment with the different options as long as they don't affect the results.
8. Run the command go f1 foo where f1.c is the name of the file containing the evaluation function (don't include the ".c" in the go command) and foo is the file suffix specified in the setup program. This will build your evaluation function, run the code, and generate the report. Your report will be in file rep.foo where foo is the file suffix specified in the setup program.
9. The setup program should be run at least once for each new evaluation function, but if you want to tweak with individual parameters, it will be quicker to modify in.foo directly instead of re-running setup.

• In between trial runs, if you change a parameter, at the end of the run it will say, "rep.foo: File exists", where foo is the extension that you specified in the setup program, and will not generate a new report. In order to generate a new report, you must either remove (rm) the report file before the run, or run the command rm rep.foo; report foo > rep.foo after the run, where foo is the extension you chose in the setup step.
• If you set the report interval too low and the number of trial runs too high, the software won't print any data to the report file. Set the report interval to the population size to get a report for every generation.
• The "M" option (make sure it is capitalized) maximizes the evaluation (fitness) function. The default is to minimize it.
• After the run, the file min.foo, where foo is the extension you chose in the setup step, will contain the best strings from the runs. The number of chromosomes in this file is the number specified for "Structures Saved" during the setup. A line in this file looks as follows:

  101111111111111111111111111111 1.0000e+00 145 4362

  From left to right, this is the actual chromosome, the fitness value, the generation that the structure first appeared, and the trial count that the structure first appeared.
• To combine a bunch of reports into one file, do the following:
  
  • Use a graphical text editor on UNIX (e.g., xemacs), or WS_FTP your files back to Windows and arrange everything there.
  • If you're a UNIX buff, run the following commands:
    
    1. touch newreportfile, where newreportfile is the name of the new file that will contain your merged reports. This will create a file of size 0 named "newreportfile".
    2. For each report file that you have, type cat reportfile >> newreportfile. This will append the contents of reportfile onto the end of newreportfile without overwriting the file.