def multi_species_benchmarks_opt2(lamR):
repeats = 50
lamP = 1
#lamSs = [0.3,0.5,0.7,1]#+list(np.logspace(-1.5, 1.5, 3))
#lamRs = [3,5,7]#np.logspace(0, 2, 5)
iron_conds = range(ba.iron_conds)
timeseries_conds = range(ba.timeseries_conds)
subtilis_conds = range(ba.subtilis_conds)
(BS_priors, sign) = ba.load_priors('gsSDnamesWithActivitySign082213','B_subtilis')
orth = ba.load_orth('bs_ba_ortho_804',['B_anthracis','B_subtilis']) #TODO make the order impossible to get
lamR = lamR[0]
lamS = 0
acc_s = 0
acc_a = 0
aupr = 0
corr = 0
for rep in range(repeats):
random.shuffle(iron_conds)
random.shuffle(timeseries_conds)
random.shuffle(subtilis_conds)
sub_s_i = int(len(subtilis_conds)/8)
sub_i_i = int(len(iron_conds)*0.75)
sub_t_i = int(len(timeseries_conds)*0.75)
sub_s = subtilis_conds[0:sub_s_i]
sub_i = iron_conds[0:sub_i_i]
sub_t = timeseries_conds[0:sub_t_i]
ba.run_both_adjust(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp2', sub_s=sub_s, sub_i=sub_i, sub_t=sub_t)
sub_st = subtilis_conds[sub_s_i:]
sub_it = iron_conds[sub_i_i:]
sub_tt = timeseries_conds[sub_t_i:]
(bs_e, bs_t, bs_genes, bs_tfs) = ba.load_B_subtilis(sub_st)
(ba_e, ba_t, ba_genes, ba_tfs) = ba.load_B_anthracis(sub_it, sub_tt)
corr += ba.betas_fused_corr('tmp2_subtilis', 'tmp2_anthracis', orth)
print corr/(rep+1)
eval_s = ba.eval_prediction('tmp2_subtilis', bs_e, bs_t, bs_genes, bs_tfs,'R2')
eval_a = ba.eval_prediction('tmp2_anthracis', ba_e, ba_t, ba_genes, ba_tfs,'R2')
(net, bs_genes, bs_tfs) = ba.load_network('tmp2_subtilis')
aupr += ba.eval_network_pr(net, bs_genes, bs_tfs, BS_priors)
print aupr/(rep+1)
acc_s += eval_s
acc_a += eval_a
return aupr/repeats
def single_species_benchmarks():
repeats = 2
lamPs = np.array([0])#np.logspace(0, 3, 10)
lamSs = np.array([0])
lamRs = np.logspace(0, 3, 5)
iron_conds = range(ba.iron_conds)
timeseries_conds = range(ba.timeseries_conds)
subtilis_conds = range(ba.subtilis_conds)
outf = file('benchmark_results/single_species_bench','w')
for lamP in lamPs:
for lamS in lamSs:
for lamR in lamRs:
acc_s = 0
acc_a = 0
for rep in range(repeats):
random.shuffle(iron_conds)
random.shuffle(timeseries_conds)
random.shuffle(subtilis_conds)
sub_s = subtilis_conds[0:int(len(subtilis_conds)/2)]
sub_i = iron_conds[0:int(len(iron_conds)/2)]
sub_t = timeseries_conds[0:int(len(timeseries_conds)/2)]
ba.run_both(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp', sub_s = sub_s, sub_i = sub_i, sub_t = sub_t)
sub_st = subtilis_conds[int(len(subtilis_conds)/2):len(subtilis_conds)]
sub_it = iron_conds[int(len(iron_conds)/2):len(iron_conds)]
sub_tt = timeseries_conds[int(len(timeseries_conds)/2):len(timeseries_conds)]
(bs_e, bs_t, bs_genes, bs_tfs) = ba.load_B_subtilis(sub_st)
(ba_e, ba_t, ba_genes, ba_tfs) = ba.load_B_anthracis(sub_it, sub_tt)
eval_s = ba.eval_prediction('tmp_subtilis', bs_e, bs_t, bs_genes, bs_tfs,'R2')
eval_a = ba.eval_prediction('tmp_anthracis', ba_e, ba_t, ba_genes, ba_tfs,'R2')
acc_s += eval_s
acc_a += eval_a
writestr = 'lamP=%f\tlamS=%f\tlamR=%f\tsubtilis=%f\tanthracis=%f\n' % (lamP, lamS, lamR, acc_s/repeats, acc_a/repeats)
outf.write(writestr)
print writestr
outf.close()
def multi_species_benchmarks2(solver):
repeats = 50
lamPs = np.array([1])#np.logspace(0, 3, 10)
lamSs = [0]#+list(np.logspace(-1.5, 1.5, 3))
lamRs = [3]#np.logspace(0, 2, 5)
iron_conds = range(ba.iron_conds)
timeseries_conds = range(ba.timeseries_conds)
subtilis_conds = range(ba.subtilis_conds)
outf = file('benchmark_results/multi_species_bench4','w')
(BS_priors, sign) = ba.load_priors('gsSDnamesWithActivitySign082213','B_subtilis')
orth = ba.load_orth('bs_ba_ortho_804',['B_anthracis','B_subtilis']) #TODO make the order impossible to get
for lamP in lamPs:
for lamS in lamSs:
for lamR in lamRs:
acc_s = 0
acc_a = 0
aupr = 0
corr = 0
for rep in range(repeats):
random.shuffle(iron_conds)
random.shuffle(timeseries_conds)
random.shuffle(subtilis_conds)
sub_s_i = int(len(subtilis_conds)/5)
sub_i_i = int(len(iron_conds)*0.75)
sub_t_i = int(len(timeseries_conds)*0.75)
sub_s = subtilis_conds[0:sub_s_i]
sub_i = iron_conds[0:sub_i_i]
sub_t = timeseries_conds[0:sub_t_i]
if solver=='run_both':
ba.run_both(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp2', sub_s = sub_s, sub_i = sub_i, sub_t = sub_t)
if solver=='run_both_adjust':
ba.run_both_adjust(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp2', sub_s=sub_s, sub_i=sub_i, sub_t=sub_t)
if solver=='run_both_refit':
ba.run_both_refit(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp2', sub_s = sub_s, sub_i = sub_i, sub_t = sub_t, it=10,k=20)
if solver=='scad':
ba.run_both(lamP=lamP, lamR=lamR, lamS=lamS, outf='tmp2', sub_s=sub_s, sub_i=sub_i, sub_t=sub_t, it=10, k=20, it_s=10)
sub_st = subtilis_conds[sub_s_i:]
sub_it = iron_conds[sub_i_i:]
sub_tt = timeseries_conds[sub_t_i:]
(bs_e, bs_t, bs_genes, bs_tfs) = ba.load_B_subtilis(sub_st)
(ba_e, ba_t, ba_genes, ba_tfs) = ba.load_B_anthracis(sub_it, sub_tt)
corr += ba.betas_fused_corr('tmp2_subtilis', 'tmp2_anthracis', orth)
print corr/rep
eval_s = ba.eval_prediction('tmp2_subtilis', bs_e, bs_t, bs_genes, bs_tfs,'R2')
eval_a = ba.eval_prediction('tmp2_anthracis', ba_e, ba_t, ba_genes, ba_tfs,'R2')
(net, bs_genes, bs_tfs) = ba.load_network('tmp2_subtilis')
aupr += ba.eval_network_pr(net, bs_genes, bs_tfs, BS_priors)
print aupr/rep
acc_s += eval_s
acc_a += eval_a
writestr = 'lamP=%f\tlamS=%f\tlamR=%f\tsubtilis=%f\tanthracis=%f\taupr=%f\tcorr_bs=%f\n' % (lamP, lamS, lamR, acc_s/repeats, acc_a/repeats, aupr/repeats, corr/repeats)
outf.write(writestr)
print writestr
outf.close()