TOT_PATHS = 150 TOT_ORGS = 100 GENES_PER_PATH = 15 PATHS_PER_ORG = 10 ORGS_PER_POP = 50 POPS = 10 SAMPLE_SIZE = int(1.5*TOT_GENES) RUNS = 10 for run in range(RUNS): print run pops = make_pops.make_testable_pops(num_genes = TOT_GENES, num_pathways = TOT_PATHS, num_orgs = TOT_ORGS, genes_per_pathway = GENES_PER_PATH, pathways_per_org = PATHS_PER_ORG, orgs_per_pop = ORGS_PER_POP, num_pops = POPS) matrix = sample_pops.construct_network_matrix(pops, SAMPLE_SIZE) construct_matrix.save_connection_matrix("data_standardruns//data%s.dat" % \ str(run).rjust(2,'0'), matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])
reload(construct_matrix) TOT_GENES = 500 TOT_PATHS = 150 TOT_ORGS = 100 GENES_PER_PATH = 15 PATHS_PER_ORG = 10 ORGS_PER_POP = 50 POPS = 50 SAMPLE_SIZE = int(1.5 * TOT_GENES) RUNS = 10 for run in range(RUNS): print run pops = make_pops.make_testable_pops(num_genes=TOT_GENES, num_pathways=TOT_PATHS, num_orgs=TOT_ORGS, genes_per_pathway=GENES_PER_PATH, pathways_per_org=PATHS_PER_ORG, orgs_per_pop=ORGS_PER_POP, num_pops=POPS) matrix = sample_pops.construct_network_matrix(pops, SAMPLE_SIZE) construct_matrix.save_connection_matrix("data_standardruns//data%s.dat" % \ str(run).rjust(2,'0'), matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])
reload(construct_matrix) TOT_GENES = 1500 TOT_PATHS = 250 TOT_ORGS = 300 GENES_PER_PATH = 15 PATHS_PER_ORG = 50 ORGS_PER_POP = 200 POPS = 10 SAMPLE_SIZE = 1.5 * TOT_GENES RUNS = 1 for run in range(10): print sample_size pops = make_pops.make_testable_pops(num_genes=TOT_GENES, num_pathways=TOT_PATHS, num_orgs=TOT_ORGS, genes_per_pathway=GENES_PER_PATH, pathways_per_org=PATHS_PER_ORG, orgs_per_pop=ORGS_PER_POP, num_pops=POPS) matrix = sample_pops.construct_network_matrix(pops, SAMPLE_SIZE) construct_matrix.save_connection_matrix( "data_standardruns//data%02d.dat" % run, matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])
TOT_GENES = 1500 TOT_PATHS = 250 TOT_ORGS = 300 GENES_PER_PATH = 15 PATHS_PER_ORG = 50 ORGS_PER_POP = 200 POPS = 10 SAMPLE_SIZE = 1.5 * TOT_GENES RUNS = 1 for run in range(10): print sample_size pops = make_pops.make_testable_pops( num_genes=TOT_GENES, num_pathways=TOT_PATHS, num_orgs=TOT_ORGS, genes_per_pathway=GENES_PER_PATH, pathways_per_org=PATHS_PER_ORG, orgs_per_pop=ORGS_PER_POP, num_pops=POPS, ) matrix = sample_pops.construct_network_matrix(pops, SAMPLE_SIZE) construct_matrix.save_connection_matrix("data_standardruns//data%02d.dat" % run, matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])
TOT_GENES = 1500 TOT_PATHS = 250 TOT_ORGS = 300 GENES_PER_PATH = 15 PATHS_PER_ORG = 50 ORGS_PER_POP = 200 POPS = 10 RUNS = 2 for run in [1]: pops = make_pops.make_testable_pops(num_genes=TOT_GENES, num_pathways=TOT_PATHS, num_orgs=TOT_ORGS, genes_per_pathway=GENES_PER_PATH, pathways_per_org=PATHS_PER_ORG, orgs_per_pop=ORGS_PER_POP, num_pops=POPS) for sample_size in [100, 500, 1000, 1500, 2250, 3000, 9999]: print sample_size matrix = sample_pops.construct_network_matrix(pops, sample_size) construct_matrix.save_connection_matrix("data_samplesize//data%s_%s.dat"\ % (str(sample_size).rjust(4,'0'), str(run).rjust(2, '0')), matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])
TOT_PATHS = 250 TOT_ORGS = 300 GENES_PER_PATH = 15 PATHS_PER_ORG = 50 ORGS_PER_POP = 200 POPS = 10 RUNS = 2 for run in [1]: pops = make_pops.make_testable_pops(num_genes = TOT_GENES, num_pathways = TOT_PATHS, num_orgs = TOT_ORGS, genes_per_pathway = GENES_PER_PATH, pathways_per_org = PATHS_PER_ORG, orgs_per_pop = ORGS_PER_POP, num_pops = POPS) for sample_size in [100, 500, 1000, 1500, 2250, 3000, 9999]: print sample_size matrix = sample_pops.construct_network_matrix(pops, sample_size) construct_matrix.save_connection_matrix("data_samplesize//data%s_%s.dat"\ % (str(sample_size).rjust(4,'0'), str(run).rjust(2, '0')), matrix) ## print(make_pops.Pathway.extant_pathways.values()[0]) ## for pop in pops: ## print(make_pops.Population.extant_populations[pop.id]) ## print(make_pops.Organism.extant_organisms[pop.id]) ## print(make_pops.Pathway.extant_pathways[pop.id])