Output Files

ATHENA produces 3 types of files during its run: summary, logs and plots. The files will have the name specified in the --out (OUT) parameter which can include a full path. Each will have an extension added to the base name as described below.

Summary file

ATHENA writes a single summary file for each run it performs. It displays the best network for each cross-validation in the run. The file will have "_summary.txt" appended to the base name specified in the parameters (--out).

The file constists of 3 sections:

cross-validation results

The top portion is a tab delimited table of the results across the cross-validations in the ATHENA run.

CV  Variables   r-squared   Training    Testing   Training-missing  Testing-missing
1   SNP9-a SNP25-b  0.0934  0.087   0.00%   0.00%
2   SNP41-a SNP53-a 0.1237  0.0598  0.00%   0.00%
3   SNP17-a C33 0.0974  0.0345    0.00%    0.00%
4   SNP57-b SNP43-b SNP72-a SNP62-b SNP40-a 0.0337  -0.0503 0.00%   0.00%
5   C48 SNP12-a 0.0874  0.0596   0.00%  0.00%

The columns are:

CV	cross-validation interval
Variables	Variables appearing in the best evolved network for the cross-vaidation
[r-squared or balacc] Training	Training fitness for the network
[r-squared or balacc] Testing	Testing fitness for the network
Training-missing	Percentage of missing data for the network on the testing data
Testing-missing	Percentage of missing data for the network on the testing data

best networks

The second section of the summary file displays the best networks in each cross-validation. These models are simplified by calculating the constant values in the models.

 CV      Model
 1       PD([(3.20 * SNP9-a),(85.16 * SNP25-b)])
 2       PM([(0.48 * SNP41-a),(0.60 * SNP53-a)])
 3       PM([(0.28 * SNP17-a),(0.93 * C33)])
 4       PS([(0.53 * PS([(0.66 * SNP57-b),(0.10 * PD([(24.42 * SNP43-b),(0.56 * SNP72-a)]))])),(0.75 * PM([(-0.47 * SNP62-b),(0.53 * SNP40-a)]))])
 5       PS([(0.25 * C48),(-0.26 * SNP12-a)])

original network representation

The final section of the summary file displays the best networks in each cross-validation as originally evolved by ATHENA. These models display the internal representation of the networks during a run.

CV      Model
1       PD([((float(8.9) - float(5.7)) * x[16]),(float(85.16) * x[49])])
2       PM([(float(.48) * x[80]),(pdiv(float(0.5),float(.83)) * x[104])])
3       PM([(float(.28) * x[32]),(float(0.93) * x[232])])
4       PS([(float(.53) * PS([(float(0.66) * x[113]),(float(.1) * PD([(float(24.42) * x[85]),(float(.56) * x[142])]))])),(float(.75) * PM([((pdiv(((float(.3) + float(.6)) * pdiv(float(.5),float(.8))),float(4.2)) - float(.6)) * x[123]),(float(.53) * x[78])]))])
5       PS([(float(.25) * x[247]),((float(.89) * (float(.6) - float(.89))) * x[22])])

Log files

The ATHENA log files are tab-delmited files with the ".log" extension. ATHENA writes one file for each cross-validation interval in the run and the cross-validation number is indicate in the filename (i.e. athena.cv1.log). These files can be used to analyze the performance of the algorithm over the course of a run.

The header line identifies each column:

gen	Generation
invalid	Number of invalid individuals in the population
avg	Training fitness average for all valid individuals
std	Training fitness standard deviation
min	Training fitness minimum score
max	Training fitness maximum score
fitness_test	Testing fitness for best individual (only calculated on last generation)
best_ind_length	Genome size for best individdual in population
avg_length	Average genome size for all individuals in population
best_ind_nodes	Number of mapped nodes in tree for best individual
avg_nodes	Average number of mapped nodes in tree for all individuals
best_ind_depth	Depth of mapped tree for best individual in population
avg_depth	Average depth of mapped trees for all individuals in population
avg_used_codons	Average number of codons used in mapping
best_ind_used_codons	Number of codons used in mapping best individual
structural_diversity	Fraction of individuals producing unique structures
selection_time	Time spent in selection process
generation_time	Time spent for entire generation
best_phenotype	Best network (by trainig score) in this generation

In the case of a parallelized run, the log files contain the combined values from all the individual populations.

Plot files

ATHENA utilizes the netgraph Python package to generate plots of the best individuals in each cross-validation. The files are in the png format and have filename extensions indicating the cross-validation (i.e. athena.cv1.png). Specific colors can be used to designate the different input variables in the Color Mapping File.