def plot_rocs(title, rocs, num_pssms):
"""
Plot roc graphs.
"""
import pylab as P
for bg_type, rocs_for_bg in rocs.iteritems():
for dataset, rocs_for_dataset in rocs_for_bg.iteritems():
tag = '%s-%s-%s' % (title, bg_type, dataset)
fig = P.figure()
for p in num_pssms:
plot_roc_points(rocs_for_dataset[p][::-1], label='%d PSSMs' % (p+1))
P.legend(loc='lower right')
P.title(tag)
plot_random_classifier()
label_plot()
P.savefig(os.path.join(options.output_dir, 'roc-%s.png' % tag))
P.close(fig)
def plot_param_setting_rocs(test_cases, use_cumulative_dists, use_p_value, use_score, label, color, linestyle):
"Test a particular setting of the biopsy module's PSSM parameters."
B.PssmParameters.use_cumulative_dists = use_cumulative_dists
B.PssmParameters.use_p_value = use_p_value
B.PssmParameters.use_score = use_score
start_time = time.time()
positive_thresholds, negative_thresholds = thresholds_for_classifier(test_cases, BiFaClassifier())
elapsed = time.time() - start_time
logging.info(
'Testing parameter settings: %40s : %10.3f : %10.1f secs',
label,
sum(positive_thresholds) + sum(negative_thresholds),
elapsed
)
rocs = roc.rocs_from_thresholds(positive_thresholds, negative_thresholds, num_points=100)
roc.plot_roc_points(rocs, label=label, color=color, ls=linestyle)
return positive_thresholds, negative_thresholds
def plot_rocs(rocs, filename, fig_title):
from pylab import figure, close, savefig, plot, title, xlabel, ylabel
figure()
plot_roc_points(
rocs,
marker='s',
color='black',
linestyle='--'
)
plot(
[0,1],
[0,1],
color='black',
linestyle=':'
)
title(fig_title)
xlabel('1 - specificity: 1-TN/(TN+FP)')
ylabel('sensitivity: TP/(TP+FN)')
logging.info('Saving ROC to %s' % filename)
savefig(filename)
(method, R.area_under_curve([roc for roc, t in roc_thresholds[method]]))
for method in methods
)
aucs[bg]['AUC50'] = dict(
(method, auc50s[method][0])
for method in methods
)
# ROC curves
P.figure()
for method in methods:
rocs = [roc for roc, t in roc_thresholds[method]]
auc = aucs[bg]['AUC'][method]
auc50 = aucs[bg]['AUC50'][method]
R.plot_roc_points(rocs, label=method, marker=markers[method], color=colors[method])
R.plot_random_classifier(label='Random')
R.label_plot()
P.legend(loc='lower right')
P.title('%s - %s' % (fragment_name(fragment), bg))
P.savefig(os.path.join(options.results_dir, 'ROC-%s-%s.png' % (fragment, bg)))
P.savefig(os.path.join(options.results_dir, 'ROC-%s-%s.eps' % (fragment, bg)))
P.close()
# precision-recall curves
P.figure()
for method in methods:
rocs = [roc for roc, t in roc_thresholds[method]]
R.plot_precision_versus_recall(rocs, label=method, marker=markers[method], color=colors[method])
R.label_precision_versus_recall()
methods = []
datasets_to_ignore = set(('T99004',))
show_title = False
sys.argv.pop(0) # remove script name
while sys.argv:
method = sys.argv.pop(0)
methods.append(method)
file_glob = sys.argv.pop(0)
logging.info('Method: %s; glob: %s', method, file_glob)
load_stat_files(glob(file_glob), overall_stats[method])
styles = [ '-', '--' ]
for bg in bg_types:
for dataset in datasets:
P.figure()
for style, method in zip(styles, methods):
stats = overall_stats[method]
rocs = list(stats[bg][0][dataset].iteritems())
rocs.sort()
rocs = [r[1] for r in rocs]
plot_roc_points(rocs, linestyle=style, color='k', label=method)
label_plot()
P.legend(loc='lower right')
if show_title:
P.title('%s - %s' % (dataset, bg))
plot_random_classifier(label='random')
P.savefig('ROC-%s-%s.png' % (bg, dataset))
P.savefig('ROC-%s-%s.eps' % (bg, dataset))
P.clf()
if 'GLAM2-i7' == method:
return 'GLAM2'
return method
for bg in backgrounds:
# ROC curves
P.figure()
for method in methods:
rocs = R.picked_rocs_from_thresholds(
scores[(method,)],
scores[(method, bg)],
num_points=options.num_points
)
auc = R.area_under_curve(rocs)
R.plot_roc_points(rocs, label='%.2f %s'%(auc,name(method)), marker=markers[method])
R.plot_random_classifier(label='0.50 Random')
R.label_plot()
P.legend(loc='lower right')
P.title('Full Sp1 - %s' % bg)
P.savefig('ROC-Sp1-%s.png' % bg)
P.savefig('ROC-Sp1-%s.eps' % bg)
# precision-recall curves
P.figure()
for method in methods:
rocs = R.picked_rocs_from_thresholds(
scores[(method,)],
scores[(method, bg)],
num_points=options.num_points
)
def do_pwm(tag, pwm, color, marker):
roc_points = calc_or_unpickle_roc(tag, pwm, positive_seqs, negative_seqs)
roc.plot_roc_points(roc_points, label=tag, color=color, marker=marker)
methods = []
datasets_to_ignore = set(('T99004', ))
show_title = False
sys.argv.pop(0) # remove script name
while sys.argv:
method = sys.argv.pop(0)
methods.append(method)
file_glob = sys.argv.pop(0)
logging.info('Method: %s; glob: %s', method, file_glob)
load_stat_files(glob(file_glob), overall_stats[method])
styles = ['-', '--']
for bg in bg_types:
for dataset in datasets:
P.figure()
for style, method in zip(styles, methods):
stats = overall_stats[method]
rocs = list(stats[bg][0][dataset].iteritems())
rocs.sort()
rocs = [r[1] for r in rocs]
plot_roc_points(rocs, linestyle=style, color='k', label=method)
label_plot()
P.legend(loc='lower right')
if show_title:
P.title('%s - %s' % (dataset, bg))
plot_random_classifier(label='random')
P.savefig('ROC-%s-%s.png' % (bg, dataset))
P.savefig('ROC-%s-%s.eps' % (bg, dataset))
P.clf()
if 'GLAM2-i7' == method:
return 'GLAM2'
return method
for bg in backgrounds:
# ROC curves
P.figure()
for method in methods:
rocs = R.picked_rocs_from_thresholds(scores[(method, )],
scores[(method, bg)],
num_points=options.num_points)
auc = R.area_under_curve(rocs)
R.plot_roc_points(rocs,
label='%.2f %s' % (auc, name(method)),
marker=markers[method])
R.plot_random_classifier(label='0.50 Random')
R.label_plot()
P.legend(loc='lower right')
P.title('Full Sp1 - %s' % bg)
P.savefig('ROC-Sp1-%s.png' % bg)
P.savefig('ROC-Sp1-%s.eps' % bg)
# precision-recall curves
P.figure()
for method in methods:
rocs = R.picked_rocs_from_thresholds(scores[(method, )],
scores[(method, bg)],
num_points=options.num_points)
R.plot_precision_versus_recall(rocs,
