def create_top_correlations_html(data, fitter, fits, scores, regions, n_top=None):
if n_top is None:
n_top = len(scores)
basedir = join(results_dir(), fit_results_relative_path(data,fitter))
ensure_dir(basedir)
gene_dir = 'gene-subplot'
series_dir = 'gene-region-fits'
def key_func(score):
g,r,pval,lst_R2 = score
return r
scores.sort(key=key_func)
top_genes = [g for g,r,pval,lst_R2 in scores[:n_top]]
top_scores = {g:r for g,r,pval,lst_R2 in scores[:n_top]}
top_pvals = {g:pval for g,r,pval,lst_R2 in scores[:n_top]}
def get_onset_time(fit):
a,h,mu,_ = fit.theta
age = age_scaler.unscale(mu)
txt = 'onset = {:.3g} years'.format(age)
cls = ''
return txt,cls
create_html(
data, fitter, fits, basedir, gene_dir, series_dir,
gene_names = top_genes,
region_names = regions,
extra_columns = [('r',top_scores),('p-value',top_pvals)],
extra_fields_per_fit = [get_onset_time],
b_inline_images = True,
b_R2_dist = False,
ttl = 'Fit for genes with top Spearman correlations',
filename = 'top-gradual-maturation',
)
def flush():
prints = []
for name, vals in _since_last_flush.items():
prints.append("{}\t{}".format(name, np.mean(list(vals.values()))))
_since_beginning[name].update(vals)
#print(_since_beginning[name])
x_vals = np.sort(list(_since_beginning[name].keys()))
y_vals = [_since_beginning[name][x] for x in x_vals]
plt.clf()
plt.plot(x_vals, y_vals)
plt.xlabel('iteration')
plt.ylabel(name)
fpath = os.path.join('debug', name.replace(' ', '_') + '.png')
base_dir = os.path.dirname(fpath)
ensure_dir(base_dir)
plt.savefig(fpath)
print("iter {}\t{}".format(_iter[0], "\t".join(prints)))
_since_last_flush.clear()
with open('debug/log.pkl', 'wb') as f:
pickle.dump(dict(_since_beginning), f, -1) #pickle.HIGHEST_PROTOCOL)
def plot_and_save_all_genes(data, fitter, fits, dirname,
show_change_distributions):
ensure_dir(dirname)
to_plot = []
genes = set(
) # use the genes from the fits and not from 'data' to support sharding (k_of_n)
for ds_fits in fits.itervalues():
for g, r in ds_fits.iterkeys():
genes.add(g)
for g in sorted(genes):
genedir = join(dirname, g[:g.index(cfg.exon_separator)]
) #does nothing if gene name does not contain the exon
ensure_dir(genedir)
filename = join(genedir, '{}.png'.format(g))
if isfile(filename):
print 'Figure already exists for gene {}. skipping...'.format(g)
continue
region_series_fits = _extract_gene_data(data, g, fits)
if show_change_distributions:
bin_centers = fits.change_distribution_params.bin_centers
else:
bin_centers = None
to_plot.append((g, region_series_fits, filename, bin_centers))
pool = Parallel(_plot_genes_job)
pool(pool.delay(*args) for args in to_plot)
def generate_image(self, training_iter):
samples = self.sess.run(self.fixed_noise_samples)
debug_dir = self.checkpoint_dir.replace('output', 'debug')
ensure_dir(debug_dir)
tflib.save_images.save_images(
(samples.reshape((len(samples), 64, 64, 3)) + 1) / (2.0),
os.path.join(debug_dir, 'samples_{}.png'.format(training_iter)))
def save_file(filename, lines):
lines.append('')
txt = '\n'.join(lines)
ensure_dir(dirname(filename))
print 'Saving to {}'.format(filename)
with open(filename, 'w') as f:
f.write(txt)
def sendToDetection(frame):
'''
sendToDetection(frame)
---
'''
ensure_dir(pic_dir) # check if dir exists
tmp_pic = pic_dir + 'pic_' + '0' +'.jpg'
detected = []
detected_faces = []
cv2.imwrite(tmp_pic, frame) # Saving frame for a while
try:
# time.sleep(wait_time)
detected = CF.face.detect(tmp_pic)
except Exception as e:
print(colored(e, color='grey'))
print('#######################################################')
print(colored(' Detection...' + spc(31) + '[ ' , color='white') + colored('OK', color='green') + colored(' ]' , color='white'))
for detected_face in detected:
detected_faces.append({'faceId' : detected_face['faceId'], 'faceRectangle': detected_face['faceRectangle']}) # Remembering face IDs and rectangles
print(colored(' Face detected: ', color='white') + colored(detected_face['faceId'], color='yellow'))
print('#######################################################')
return detected_faces
def _consolidate(dct_res, base_filename, k_of_n, found_keys_not_in_main_file):
filename = _cache_filename(base_filename, k_of_n)
# write the updated main file
if found_keys_not_in_main_file:
if cfg.verbosity > 0:
print 'Writing back consolidated fit file...'
ensure_dir(dirname(filename))
with open(filename,'w') as f:
pickle.dump(dct_res,f)
if cfg.verbosity > 0:
print 'Deleting any partial fit files...'
if k_of_n is None:
# it's the main file - delete all k_of_n files and the batches dir
batchdir = _batch_dir(base_filename)
if isdir(batchdir):
shutil.rmtree(batchdir)
partial_files = set(glob(filename + '*')) - {filename}
for filename in partial_files:
os.remove(filename)
else:
# it's a shard - delete just the batches for that file
base = _batch_base_filename(base_filename,k_of_n)
batch_filenames = glob(base + '*')
for filename in batch_filenames:
os.remove(filename)
def save_file(filename, lines):
lines.append("")
txt = "\n".join(lines)
ensure_dir(dirname(filename))
print "Saving to {}".format(filename)
with open(filename, "w") as f:
f.write(txt)
def _consolidate(dct_res, base_filename, k_of_n, found_keys_not_in_main_file):
filename = _cache_filename(base_filename, k_of_n)
# write the updated main file
if found_keys_not_in_main_file:
if cfg.verbosity > 0:
print 'Writing back consolidated fit file...'
ensure_dir(dirname(filename))
with open(filename, 'w') as f:
pickle.dump(dct_res, f)
if cfg.verbosity > 0:
print 'Deleting any partial fit files...'
if k_of_n is None:
# it's the main file - delete all k_of_n files and the batches dir
batchdir = _batch_dir(base_filename)
if isdir(batchdir):
shutil.rmtree(batchdir)
partial_files = set(glob(filename + '*')) - {filename}
for filename in partial_files:
os.remove(filename)
else:
# it's a shard - delete just the batches for that file
base = _batch_base_filename(base_filename, k_of_n)
batch_filenames = glob(base + '*')
for filename in batch_filenames:
os.remove(filename)
def test_batch(self):
"""Tests the image batch generator."""
output_dir = os.path.join(self.debug_dir, 'test_batch')
ensure_dir(output_dir)
img, target = self.train_data_gen().next()
img = img.reshape([self.batch_size] + self.image_dim)
save_images_files(img / 255.0, output_dir=output_dir, labels=target)
def create_top_genes_html(data, fitter, fits, scores, regions, n_top=None, filename_suffix=''):
if n_top is None:
n_top = len(scores)
basedir = join(results_dir(), fit_results_relative_path(data,fitter))
ensure_dir(basedir)
gene_dir = 'gene-subplot'
series_dir = 'gene-region-fits'
def key_func(score):
g,pval,qval = score
return pval
scores.sort(key=key_func)
top_genes = [g for g,pval,qval in scores[:n_top]]
top_pvals = {g:pval for g,pval,qval in scores[:n_top]}
top_qvals = {g:qval for g,pval,qval in scores[:n_top]}
n = len(scores)
n05 = len([g for g,pval,qval in scores if qval < 0.05])
n01 = len([g for g,pval,qval in scores if qval < 0.01])
top_text = """\
<pre>
one sided t-test: {regions[0]} < {regions[1]}
{n05}/{n} q-values < 0.05
{n01}/{n} q_values < 0.01
</pre>
""".format(**locals())
def get_onset_time(fit):
a,h,mu,_ = fit.theta
age = age_scaler.unscale(mu)
return 'onset = {:.3g} years'.format(age)
def get_onset_dist(fit):
mu_vals = fit.theta_samples[2,:]
mu = mu_vals.mean()
vLow,vHigh = np.percentile(mu_vals, (20,80))
mu = age_scaler.unscale(mu)
vLow = age_scaler.unscale(vLow)
vHigh = age_scaler.unscale(vHigh)
txt = 'onset reestimate (mean [20%, 80%]) = {:.3g} [{:.3g},{:.3g}]'.format(mu,vLow,vHigh)
cls = ''
return txt,cls
create_html(
data, fitter, fits, basedir, gene_dir, series_dir,
gene_names = top_genes,
region_names = regions,
extra_columns = [('p-value',top_pvals), ('q-value',top_qvals)],
extra_fields_per_fit = [get_onset_time, get_onset_dist],
b_inline_images = True,
inline_image_size = '30%',
b_R2_dist = False,
ttl = 'Fit for genes with top t-test scores',
top_text = top_text,
filename = 'gradual-maturation-t-test' + filename_suffix,
)
def do_gene_fits(data, gene, fitter, filename, b_show):
fig = plot_gene(data,gene)
if filename is None:
ensure_dir(results_dir())
filename = join(results_dir(), 'fit.png')
print 'Saving figure to {}'.format(filename)
save_figure(fig, filename)
if b_show:
plt.show(block=True)
def _save_batch(dct_updates, base_filename, k_of_n, i):
filename = _batch_base_filename(base_filename, k_of_n) + str(i)
ensure_dir(dirname(filename))
# if there's already a file by that name, merge its contents
if isfile(filename):
dct_existing = _read_one_cache_file(filename, st_keys=None, is_batch=True)
dct_updates.update(dct_existing)
with open(filename,'w') as f:
pickle.dump(dct_updates,f)
def main(cfg, argv=None):
FLAGS = tf.app.flags.FLAGS
GAN = dataset_gan_dict[FLAGS.dataset_name]
gan = GAN(cfg=cfg, test_mode=True)
gan.load_generator()
# Setting test time reconstruction hyper parameters.
[tr_rr, tr_lr, tr_iters] = [FLAGS.rec_rr, FLAGS.rec_lr, FLAGS.rec_iters]
if FLAGS.defense_type.lower() != 'none':
if FLAGS.rec_path and FLAGS.defense_type == 'defense_gan':
# Extract hyperparameters from reconstruction path.
if FLAGS.rec_path:
train_param_re = re.compile('recs_rr(.*)_lr(.*)_iters(.*)')
[tr_rr, tr_lr, tr_iters] = \
train_param_re.findall(FLAGS.rec_path)[0]
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
elif FLAGS.defense_type == 'defense_gan':
assert FLAGS.online_training or not FLAGS.train_on_recs
if FLAGS.override:
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
# Setting the results directory.
results_dir, result_file_name = _get_results_dir_filename(gan)
# Result file name. The counter ensures we are not overwriting the
# results.
counter = 0
temp_fp = str(counter) + '_' + result_file_name
results_dir = os.path.join(results_dir, FLAGS.results_dir)
temp_final_fp = os.path.join(results_dir, temp_fp)
while os.path.exists(temp_final_fp):
counter += 1
temp_fp = str(counter) + '_' + result_file_name
temp_final_fp = os.path.join(results_dir, temp_fp)
result_file_name = temp_fp
sub_result_path = os.path.join(results_dir, result_file_name)
accuracies = measure_gan(gan,
rec_data_path=FLAGS.rec_path,
probe_size=FLAGS.probe_size,
calc_real_data_is=FLAGS.calc_real_data_is)
ensure_dir(results_dir)
with open(sub_result_path, 'a') as f:
f.writelines([str(acc) + ' ' for acc in accuracies])
f.write('\n')
print('[*] saved accuracy in {}'.format(sub_result_path))
def do_one_fit(series, fitter, loo, filename, b_show):
if fitter is not None:
theta, sigma, LOO_predictions,_ = fitter.fit(series.ages, series.single_expression, loo=loo)
fig = plot_one_series(series, fitter.shape, theta, LOO_predictions)
else:
fig = plot_one_series(series)
if filename is None:
ensure_dir(results_dir())
filename = join(results_dir(), 'fit.png')
save_figure(fig, filename, print_filename=True)
if b_show:
plt.show(block=True)
def save(self, prefixes=None, global_step=None, checkpoint_dir=None):
if global_step is None:
global_step = self.global_step
if checkpoint_dir is None:
checkpoint_dir = self._set_checkpoint_dir
ensure_dir(checkpoint_dir)
self._initialize_saver(prefixes)
self.saver.save(self.sess,
os.path.join(checkpoint_dir, self.model_save_name),
global_step=global_step)
print('Saved at iter {} to {}'.format(self.sess.run(global_step),
checkpoint_dir))
def _save_batch(dct_updates, base_filename, k_of_n, i):
filename = _batch_base_filename(base_filename, k_of_n) + str(i)
ensure_dir(dirname(filename))
# if there's already a file by that name, merge its contents
if isfile(filename):
dct_existing = _read_one_cache_file(filename,
st_keys=None,
is_batch=True)
dct_updates.update(dct_existing)
with open(filename, 'w') as f:
pickle.dump(dct_updates, f)
def fit_serveral_genes(series, fitter, loo, filename, b_show):
if fitter is not None:
theta, L, LOO_predictions,_ = fitter.fit(series.ages, series.expression, loo=loo)
print 'L = {}'.format(L)
fig = plot_series(series, fitter.shape, theta, LOO_predictions)
else:
fig = plot_series(series)
if filename is None:
ensure_dir(results_dir())
filename = join(results_dir(), 'fits.png')
print 'Saving figure to {}'.format(filename)
save_figure(fig, filename)
if b_show:
plt.show(block=True)
def save_figure(fig, filename, b_close=False, b_square=True, show_frame=False, under_results=False, print_filename=False):
if under_results:
dirname = results_dir()
filename = join(dirname,filename)
ensure_dir(os.path.dirname(filename))
if cfg.verbosity >= 1 or print_filename:
print 'Saving figure to {}'.format(filename)
figure_size_x = cfg.default_figure_size_x_square if b_square else cfg.default_figure_size_x
fig.set_size_inches(figure_size_x, cfg.default_figure_size_y)
if show_frame:
facecolor = cfg.default_figure_facecolor
else:
facecolor = 'white'
fig.savefig(filename, facecolor=facecolor, dpi=cfg.default_figure_dpi)
if b_close:
plt.close(fig)
def fit_serveral_genes(series, fitter, loo, filename, b_show):
if fitter is not None:
theta, L, LOO_predictions, _ = fitter.fit(series.ages,
series.expression,
loo=loo)
print 'L = {}'.format(L)
fig = plot_series(series, fitter.shape, theta, LOO_predictions)
else:
fig = plot_series(series)
if filename is None:
ensure_dir(results_dir())
filename = join(results_dir(), 'fits.png')
print 'Saving figure to {}'.format(filename)
save_figure(fig, filename)
if b_show:
plt.show(block=True)
def create_top_correlations_html(data,
fitter,
fits,
scores,
regions,
n_top=None):
if n_top is None:
n_top = len(scores)
basedir = join(results_dir(), fit_results_relative_path(data, fitter))
ensure_dir(basedir)
gene_dir = 'gene-subplot'
series_dir = 'gene-region-fits'
def key_func(score):
g, r, pval, lst_R2 = score
return r
scores.sort(key=key_func)
top_genes = [g for g, r, pval, lst_R2 in scores[:n_top]]
top_scores = {g: r for g, r, pval, lst_R2 in scores[:n_top]}
top_pvals = {g: pval for g, r, pval, lst_R2 in scores[:n_top]}
def get_onset_time(fit):
a, h, mu, _ = fit.theta
age = age_scaler.unscale(mu)
txt = 'onset = {:.3g} years'.format(age)
cls = ''
return txt, cls
create_html(
data,
fitter,
fits,
basedir,
gene_dir,
series_dir,
gene_names=top_genes,
region_names=regions,
extra_columns=[('r', top_scores), ('p-value', top_pvals)],
extra_fields_per_fit=[get_onset_time],
b_inline_images=True,
b_R2_dist=False,
ttl='Fit for genes with top Spearman correlations',
filename='top-gradual-maturation',
)
def plot_and_save_all_exons_from_series(fits, exons_dir, series_dir):
to_plot = []
ensure_dir(exons_dir)
keys = set()
for ds_fits in fits.itervalues():
for g, r in ds_fits.iterkeys():
keys.add((g[:g.index(cfg.exon_separator)], r))
for g, r in keys:
gene_dir = join(exons_dir, g)
ensure_dir(gene_dir)
filename = join(gene_dir, '{}-{}.png'.format(g, r))
if isfile(filename):
print 'Figure already exists for gene {} in region {}. skipping...'.format(
g, r)
continue
to_plot.append((g, r, filename, series_dir))
pool = Parallel(_plot_exons_from_series_job)
pool(pool.delay(*args) for args in to_plot)
def build_shards(src_dir,
save_dir,
src_file,
tgt_file,
vocab,
shard_size,
feat_ext,
mode='train',
feats=None):
src_shards = split_corpus(src_file, shard_size)
tgt_shards = split_corpus(tgt_file, shard_size)
ensure_dir(save_dir)
shard_index = 0
for src_shard, tgt_shard in zip(src_shards, tgt_shards):
logger.info('Building %s shard %d' % (mode, shard_index))
audio_paths = [os.path.join(src_dir, p.strip()) for p in src_shard]
assert all([os.path.exists(p) for p in audio_paths]), \
"following audio files not found: %s" % \
' '.join([p.strip() for p in audio_paths if not os.path.exists(p)])
targets = [t.strip() for t in tgt_shard]
src_tgt_pairs = list(
zip(audio_paths, targets, cycle([feat_ext]), cycle([vocab])))
with Pool(50) as p:
result = list(
tqdm(p.imap(_worker, src_tgt_pairs), total=len(src_tgt_pairs)))
result = [r for r in result if r is not None]
audio_feats, transcriptions, indices = zip(*result)
shard = {
'src': np.asarray(audio_feats),
'tgt': np.asarray(transcriptions),
'indices':
np.asarray([np.asarray(x).reshape(-1, 1) for x in indices]),
'feats': feats
}
shard_path = os.path.join(save_dir, '%s.%05d.pt' % (mode, shard_index))
logger.info('Saving shard %d to %s' % (shard_index, shard_path))
torch.save(shard, shard_path)
shard_index += 1
def save_ds(self):
"""Reconstructs the images of the config's dataset with the
generator."""
if self.dataset_name == 'cifar':
splits = ['train', 'dev']
else:
splits = ['train', 'dev', 'test']
for split in splits:
output_dir = os.path.join('data', 'cache',
'{}_pkl'.format(self.dataset_name),
split)
if self.debug:
output_dir += '_debug'
ensure_dir(output_dir)
orig_imgs_pkl_path = os.path.join(output_dir,
'feats.pkl'.format(split))
if os.path.exists(orig_imgs_pkl_path) and not self.test_again:
with open(orig_imgs_pkl_path) as f:
all_recs = cPickle.load(f)
could_load = True
print('[#] Dataset is already saved.')
return
gen_func = getattr(self, '{}_gen_test'.format(split))
all_targets = []
orig_imgs = []
ctr = 0
for images, targets in gen_func():
ctr += 1
transformed_images = self.sess.run(self.real_data_test,
feed_dict={
self.real_data_test_pl:
images})
orig_imgs.append(transformed_images)
all_targets.append(targets)
orig_imgs = np.concatenate(orig_imgs).reshape(
[-1] + self.image_dim)
all_targets = np.concatenate(all_targets)
with open(orig_imgs_pkl_path, 'w') as f:
cPickle.dump(orig_imgs, f, cPickle.HIGHEST_PROTOCOL)
cPickle.dump(all_targets, f, cPickle.HIGHEST_PROTOCOL)
def save_fits_and_create_html(data, fitter, fits=None, basedir=None,
do_genes=True, do_series=True, do_hist=True, do_html=True, only_main_html=False,
k_of_n=None,
use_correlations=False, correlations=None,
show_change_distributions=False,
html_kw=None,
figure_kw=None):
if fits is None:
fits = get_all_fits(data,fitter,k_of_n)
if basedir is None:
basedir = join(results_dir(), fit_results_relative_path(data,fitter))
if use_correlations:
basedir = join(basedir,'with-correlations')
if html_kw is None:
html_kw = {}
if figure_kw is None:
figure_kw = {}
print 'Writing HTML under {}'.format(basedir)
ensure_dir(basedir)
gene_dir = 'gene-subplot'
series_dir = 'gene-region-fits'
correlations_dir = 'gene-correlations'
scores_dir = 'score_distributions'
if do_genes and not only_main_html: # relies on the sharding of the fits respecting gene boundaries
plot_and_save_all_genes(data, fitter, fits, join(basedir,gene_dir), show_change_distributions)
if do_series and not only_main_html:
plot_and_save_all_series(data, fitter, fits, join(basedir,series_dir), use_correlations, show_change_distributions, figure_kw)
if do_hist and k_of_n is None and not only_main_html:
create_score_distribution_html(fits, use_correlations, join(basedir,scores_dir))
if do_html and k_of_n is None:
link_to_correlation_plots = use_correlations and correlations is not None
if link_to_correlation_plots and not only_main_html:
plot_and_save_all_gene_correlations(data, correlations, join(basedir,correlations_dir))
dct_pathways = load_17_pathways_breakdown()
pathway_genes = set.union(*dct_pathways.values())
data_genes = set(data.gene_names)
missing = pathway_genes - data_genes
b_pathways = len(missing) < len(pathway_genes)/2 # simple heuristic to create pathways only if we have most of the genes (currently 61 genes are missing)
create_html(
data, fitter, fits, basedir, gene_dir, series_dir, scores_dir, correlations_dir=correlations_dir,
use_correlations=use_correlations, link_to_correlation_plots=link_to_correlation_plots,
b_pathways=b_pathways, **html_kw
)
def plot_and_save_all_genes(data, fitter, fits, dirname, show_change_distributions):
ensure_dir(dirname)
to_plot = []
genes = set() # use the genes from the fits and not from 'data' to support sharding (k_of_n)
for ds_fits in fits.itervalues():
for g,r in ds_fits.iterkeys():
genes.add(g)
for g in sorted(genes):
filename = join(dirname, '{}.png'.format(g))
if isfile(filename):
print 'Figure already exists for gene {}. skipping...'.format(g)
continue
region_series_fits = _extract_gene_data(data,g,fits)
if show_change_distributions:
bin_centers = fits.change_distribution_params.bin_centers
else:
bin_centers = None
to_plot.append((g,region_series_fits,filename, bin_centers))
pool = Parallel(_plot_genes_job)
pool(pool.delay(*args) for args in to_plot)
def get_onset_times(data, fitter, R2_threshold, b_force=False):
filename = join(cache_dir(),fit_results_relative_path(data,fitter) + '.pkl')
if isfile(filename):
print 'Loading onset distribution from {}'.format(filename)
with open(filename) as f:
bin_edges, change_vals = pickle.load(f)
else:
print 'Computing...'
fits = get_all_fits(data, fitter)
thetas = [fit.theta for fit in iterate_fits(fits, R2_threshold=R2_threshold)]
stages = [stage.scaled(age_scaler) for stage in dev_stages]
low = min(stage.from_age for stage in stages)
high = max(stage.to_age for stage in stages)
bin_edges, change_vals = compute_change_distribution(fitter.shape, thetas, low, high, n_bins=50)
print 'Saving result to {}'.format(filename)
ensure_dir(dirname(filename))
with open(filename,'w') as f:
pickle.dump((bin_edges,change_vals),f)
return bin_edges, change_vals
def create_score_distribution_html(fits, use_correlations, dirname):
ensure_dir(dirname)
with interactive(False):
hist_filename = 'R2-hist.png'
fig = plot_score_distribution(fits,use_correlations)
save_figure(fig, join(dirname,hist_filename), b_close=True)
if use_correlations:
scatter_filename = 'comparison-scatter.png'
fig = plot_score_comparison_scatter_for_correlations(fits)
save_figure(fig, join(dirname,scatter_filename), b_close=True)
delta_hist_filename = 'R2-delta-hist.png'
fig = plot_score_improvement_histogram_for_correlations(fits)
save_figure(fig, join(dirname,delta_hist_filename), b_close=True)
image_size = "50%"
from jinja2 import Template
import shutil
html = Template("""
<html>
<head>
<link rel="stylesheet" type="text/css" href="score-distribution.css">
</head>
<body>
<center><H1>R2 Score Distribution</H1></center>
<a href="{{hist_filename}}"> <img src="{{hist_filename}}" height="{{image_size}}"></a>
{% if use_correlations %}
<a href="{{delta_hist_filename}}"> <img src="{{delta_hist_filename}}" height="{{image_size}}"></a>
<a href="{{scatter_filename}}"> <img src="{{scatter_filename}}" height="{{image_size}}"></a>
{% endif %}
</body>
</html>
""").render(**locals())
filename = join(dirname,'scores.html')
with open(filename, 'w') as f:
f.write(html)
shutil.copy(join(resources_dir(),'score-distribution.css'), dirname)
def plot_and_save_all_series(data,
fitter,
fits,
dirname,
use_correlations,
show_change_distributions,
exons_layout=False,
figure_kw=None):
ensure_dir(dirname)
to_plot = []
for dsfits in fits.itervalues():
for (g, r), fit in dsfits.iteritems():
genedir = join(
dirname,
g[:g.index(cfg.exon_separator)] if exons_layout else g)
ensure_dir(genedir)
filename = join(genedir, 'fit-{}-{}.png'.format(g, r))
if isfile(filename):
print 'Figure already exists for {}@{}. skipping...'.format(
g, r)
continue
series = data.get_one_series(g, r)
if show_change_distributions and hasattr(
fit, 'change_distribution_weights'):
change_distribution = Bunch(
centers=fits.change_distribution_params.bin_centers,
weights=fit.change_distribution_weights,
)
else:
change_distribution = None
to_plot.append((series, fit, filename, use_correlations,
change_distribution, figure_kw))
if cfg.parallel_run_locally:
for args in to_plot:
_plot_series_job(*args)
else:
pool = Parallel(_plot_series_job)
pool(pool.delay(*args) for args in to_plot)
def save_figure(fig,
filename,
b_close=False,
b_square=True,
show_frame=False,
under_results=False,
print_filename=False):
if under_results:
dirname = results_dir()
filename = join(dirname, filename)
ensure_dir(os.path.dirname(filename))
if cfg.verbosity >= 1 or print_filename:
print 'Saving figure to {}'.format(filename)
figure_size_x = cfg.default_figure_size_x_square if b_square else cfg.default_figure_size_x
fig.set_size_inches(figure_size_x, cfg.default_figure_size_y)
if show_frame:
facecolor = cfg.default_figure_facecolor
else:
facecolor = 'white'
fig.savefig(filename, facecolor=facecolor, dpi=cfg.default_figure_dpi)
if b_close:
plt.close(fig)
def save_ds(gan_model):
splits = ['train', 'dev', 'test']
for split in splits:
output_dir = os.path.join('data', 'cache',
'{}_pkl'.format(gan_model.dataset_name),
split)
if gan_model.debug:
output_dir += '_debug'
ensure_dir(output_dir)
orig_imgs_pkl_path = os.path.join(output_dir,
'feats.pkl'.format(split))
if os.path.exists(orig_imgs_pkl_path) and not gan_model.test_again:
with open(orig_imgs_pkl_path) as f:
all_recs = cPickle.load(f)
could_load = True
print('[#] Dataset is already saved.')
return
gen_func = getattr(gan_model, '{}_gen_test'.format(split))
all_targets = []
orig_imgs = []
ctr = 0
for images, targets in gen_func():
ctr += 1
transformed_images = gan_model.sess.run(
gan_model.real_data_test,
feed_dict={gan_model.real_data_test_pl: images})
orig_imgs.append(transformed_images)
all_targets.append(targets)
orig_imgs = np.concatenate(orig_imgs).reshape([-1] +
gan_model.image_dim)
all_targets = np.concatenate(all_targets)
with open(orig_imgs_pkl_path, 'w') as f:
cPickle.dump(orig_imgs, f, cPickle.HIGHEST_PROTOCOL)
cPickle.dump(all_targets, f, cPickle.HIGHEST_PROTOCOL)
def plot_and_save_all_exons(data, fitter, fits, dirname):
ensure_dir(dirname)
to_plot = []
genes, regions = set(), set()
for ds_fits in fits.itervalues():
for g, r in ds_fits.iterkeys():
genes.add(g[:g.index(cfg.exon_separator)])
regions.add(r)
for g in sorted(genes):
for r in sorted(regions):
gene_dir = join(dirname, g)
ensure_dir(gene_dir)
filename = join(gene_dir, '{}-{}.png'.format(g, r))
if isfile(filename):
print 'Figure already exists for gene {} in region {}. skipping...'.format(
g, r)
continue
exons_series_fits = _extract_exons_data(data, g, r, fits)
if not np.count_nonzero(exons_series_fits):
continue
to_plot.append((g, r, exons_series_fits, filename))
pool = Parallel(_plot_exons_job)
pool(pool.delay(*args) for args in to_plot)
def _set_checkpoint_dir(self):
"""Sets the directory containing snapshots of the model."""
self.cfg_file = self.cfg['cfg_path']
if 'cfg.yml' in self.cfg_file:
ckpt_dir = os.path.dirname(self.cfg_file)
else:
ckpt_dir = os.path.join(
self.output_dir,
self.cfg_file.replace('experiments/cfgs/',
'').replace('cfg.yml',
'').replace('.yml', ''))
if not self.test_mode:
postfix = ''
ignore_list = ['dataset', 'cfg_file', 'batch_size']
if hasattr(self, 'cfg'):
if self.cfg is not None:
for prop in self.default_properties:
if prop in ignore_list:
continue
if prop.upper() in self.cfg.keys():
self_val = getattr(self, prop)
if self_val is not None:
if getattr(self,
prop) != self.cfg[prop.upper()]:
postfix += '-{}={}'.format(
prop, self_val).replace('.', '_')
ckpt_dir += postfix
ensure_dir(ckpt_dir)
self.checkpoint_dir = ckpt_dir
self.debug_dir = self.checkpoint_dir.replace('output', 'debug')
ensure_dir(self.debug_dir)
def plot_and_save_all_series(data, fitter, fits, dirname, use_correlations, show_change_distributions, figure_kw=None):
ensure_dir(dirname)
to_plot = []
for dsfits in fits.itervalues():
for (g,r),fit in dsfits.iteritems():
filename = join(dirname, 'fit-{}-{}.png'.format(g,r))
if isfile(filename):
print 'Figure already exists for {}@{}. skipping...'.format(g,r)
continue
series = data.get_one_series(g,r)
if show_change_distributions and hasattr(fit, 'change_distribution_weights'):
change_distribution = Bunch(
centers = fits.change_distribution_params.bin_centers,
weights = fit.change_distribution_weights,
)
else:
change_distribution = None
to_plot.append((series,fit,filename,use_correlations, change_distribution, figure_kw))
if cfg.parallel_run_locally:
for args in to_plot:
_plot_series_job(*args)
else:
pool = Parallel(_plot_series_job)
pool(pool.delay(*args) for args in to_plot)
def create_score_distribution_html(fits, use_correlations, dirname):
ensure_dir(dirname)
with interactive(False):
hist_filename = 'R2-hist.png'
fig = plot_score_distribution(fits, use_correlations)
save_figure(fig, join(dirname, hist_filename), b_close=True)
if use_correlations:
scatter_filename = 'comparison-scatter.png'
fig = plot_score_comparison_scatter_for_correlations(fits)
save_figure(fig, join(dirname, scatter_filename), b_close=True)
delta_hist_filename = 'R2-delta-hist.png'
fig = plot_score_improvement_histogram_for_correlations(fits)
save_figure(fig, join(dirname, delta_hist_filename), b_close=True)
image_size = "50%"
import shutil
html = get_jinja_env().get_template('R2.jinja').render(**locals())
filename = join(dirname, 'scores.html')
with open(filename, 'w') as f:
f.write(html)
shutil.copy(join(resources_dir(), 'score-distribution.css'), dirname)
def main(cfg, argv=None):
FLAGS = tf.app.flags.FLAGS
GAN = dataset_gan_dict[FLAGS.dataset_name]
gan = GAN(cfg=cfg, test_mode=True)
gan.load_generator()
# Setting test time reconstruction hyper parameters.
[tr_rr, tr_lr, tr_iters] = [FLAGS.rec_rr, FLAGS.rec_lr, FLAGS.rec_iters]
if FLAGS.defense_type.lower() != 'none':
if FLAGS.rec_path and FLAGS.defense_type == 'defense_gan':
# extract hyper parameters from reconstruction path.
if FLAGS.rec_path:
train_param_re = re.compile('recs_rr(.*)_lr(.*)_iters(.*)')
[tr_rr, tr_lr, tr_iters] = \
train_param_re.findall(FLAGS.rec_path)[0]
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
elif FLAGS.defense_type == 'defense_gan':
assert FLAGS.online_training or not FLAGS.train_on_recs
if FLAGS.override:
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
# Setting the reuslts directory
results_dir, result_file_name = _get_results_dir_filename(gan)
# Result file name. The counter makes sure we are not overwriting the
# results.
counter = 0
temp_fp = str(counter) + '_' + result_file_name
results_dir = os.path.join(results_dir, FLAGS.results_dir)
temp_final_fp = os.path.join(results_dir, temp_fp)
while os.path.exists(temp_final_fp):
counter += 1
temp_fp = str(counter) + '_' + result_file_name
temp_final_fp = os.path.join(results_dir, temp_fp)
result_file_name = temp_fp
sub_result_path = os.path.join(results_dir, result_file_name)
accuracies = blackbox(gan,
rec_data_path=FLAGS.rec_path,
batch_size=FLAGS.batch_size,
learning_rate=FLAGS.learning_rate,
nb_epochs=FLAGS.nb_epochs,
holdout=FLAGS.holdout,
data_aug=FLAGS.data_aug,
nb_epochs_s=FLAGS.nb_epochs_s,
lmbda=FLAGS.lmbda,
online_training=FLAGS.online_training,
train_on_recs=FLAGS.train_on_recs,
defense_type=FLAGS.defense_type)
ensure_dir(results_dir)
with open(sub_result_path, 'a') as f:
f.writelines([
str(accuracies[x]) + ' '
for x in ['bbox', 'sub', 'bbox_on_sub_adv_ex']
])
f.write('\n')
print('[*] saved accuracy in {}'.format(sub_result_path))
if 'roc_info' in accuracies.keys(): # For attack detection.
pkl_result_path = sub_result_path.replace('.txt', '_roc.pkl')
with open(pkl_result_path, 'w') as f:
cPickle.dump(accuracies['roc_info'], f, cPickle.HIGHEST_PROTOCOL)
print('[*] saved roc_info in {}'.format(sub_result_path))
def save_matfile(mdict, filename):
ensure_dir(dirname(filename))
print 'Saving to {}'.format(filename)
savemat(filename, mdict, oned_as='column')
def main(cfg, *args):
FLAGS = tf.app.flags.FLAGS
rng = np.random.RandomState([11, 24, 1990])
tf.set_random_seed(11241990)
gan = gan_from_config(cfg, True)
results_dir = 'results/clean/{}'.format(gan.dataset_name)
ensure_dir(results_dir)
sess = gan.sess
gan.load_model()
# use test split
train_images, train_labels, test_images, test_labels = get_cached_gan_data(
gan, test_on_dev=False, orig_data_flag=True)
x_shape = [None] + list(train_images.shape[1:])
images_pl = tf.placeholder(tf.float32,
shape=[BATCH_SIZE] +
list(train_images.shape[1:]))
labels_pl = tf.placeholder(tf.float32,
shape=[BATCH_SIZE] + [train_labels.shape[1]])
if FLAGS.num_tests > 0:
test_images = test_images[:FLAGS.num_tests]
test_labels = test_labels[:FLAGS.num_tests]
if FLAGS.num_train > 0:
train_images = train_images[:FLAGS.num_train]
train_labels = train_labels[:FLAGS.num_train]
train_params = {
'nb_epochs': 10,
'batch_size': BATCH_SIZE,
'learning_rate': 0.001
}
eval_params = {'batch_size': BATCH_SIZE}
# train classifier for mnist, fmnist
if gan.dataset_name in ['mnist', 'f-mnist']:
model = model_a(input_shape=x_shape, nb_classes=train_labels.shape[1])
preds_train = model.get_logits(images_pl, dropout=True)
model_train(sess,
images_pl,
labels_pl,
preds_train,
train_images,
train_labels,
args=train_params,
rng=rng,
init_all=False)
elif gan.dataset_name == 'cifar-10':
pre_model = Model('classifiers/model/',
tiny=False,
mode='eval',
sess=sess)
model = DefenseWrapper(pre_model, 'logits')
elif gan.dataset_name == 'celeba':
# TODO
raise NotImplementedError
model.add_rec_model(gan, batch_size=BATCH_SIZE)
preds_eval = model.get_logits(images_pl)
# calculate norms
num_dims = len(images_pl.get_shape())
avg_inds = list(range(1, num_dims))
reconstruct = gan.reconstruct(images_pl, batch_size=BATCH_SIZE)
# We use L2 loss for GD steps
diff_op = tf.reduce_mean(tf.square(reconstruct - images_pl), axis=avg_inds)
acc, mse, roc_info = model_eval_gan(sess,
images_pl,
labels_pl,
preds_eval,
None,
test_images=test_images,
test_labels=test_labels,
args=eval_params,
diff_op=diff_op)
# Logging
logfile = open(os.path.join(results_dir, 'acc.txt'), 'a+')
msg = 'lr_{}_iters_{}, {}\n'.format(gan.rec_lr, gan.rec_iters, acc)
logfile.writelines(msg)
logfile.close()
logfile = open(os.path.join(results_dir, 'mse.txt'), 'a+')
msg = 'lr_{}_iters_{}, {}\n'.format(gan.rec_lr, gan.rec_iters, mse)
logfile.writelines(msg)
logfile.close()
pickle_filename = os.path.join(
results_dir, 'roc_lr_{}_iters_{}.pkl'.format(gan.rec_lr,
gan.rec_iters))
with open(pickle_filename, 'w') as f:
cPickle.dump(roc_info, f, cPickle.HIGHEST_PROTOCOL)
print('[*] saved roc_info in {}'.format(pickle_filename))
return [acc, mse]
def save_matfile(mdict, filename):
ensure_dir(dirname(filename))
print 'Saving to {}'.format(filename)
savemat(filename, mdict, oned_as='column')
def plot_and_save_all_gene_correlations(data, correlations, dirname):
ensure_dir(dirname)
for region in data.region_names:
fig = plot_gene_correlations_single_region(correlations[region], region, data.gene_names)
save_figure(fig, join(dirname,'{}.png'.format(region)), b_close=True)
def main(cfg, argv=None):
FLAGS = tf.app.flags.FLAGS
gan = DefenseGANBase(cfg=cfg, test_mode=True)
# Setting test time reconstruction hyper parameters.
[tr_rr, tr_lr, tr_iters] = [FLAGS.rec_rr, FLAGS.rec_lr, FLAGS.rec_iters]
if FLAGS.defense_type.lower() != 'none':
if FLAGS.defense_type == 'defense_gan':
if 'GENERATOR_INIT_PATH' in cfg:
gan = DefenseGANv2(get_generator_fn(cfg['DATASET_NAME']),
cfg=cfg,
test_mode=True)
else:
gan = DefenseGANBase(cfg=cfg, test_mode=True)
gan.load_model()
# Extract hyperparameters from reconstruction path.
if FLAGS.rec_path is not None:
train_param_re = re.compile('recs_rr(.*)_lr(.*)_iters(.*)')
[tr_rr, tr_lr, tr_iters] = \
train_param_re.findall(FLAGS.rec_path)[0]
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
else:
assert FLAGS.online_training or not FLAGS.train_on_recs
if FLAGS.override:
gan.rec_rr = int(tr_rr)
gan.rec_lr = float(tr_lr)
gan.rec_iters = int(tr_iters)
# Setting the results directory.
results_dir, result_file_name = _get_results_dir_filename(gan)
# Result file name. The counter ensures we are not overwriting the
# results.
counter = 0
temp_fp = str(counter) + '_' + result_file_name
results_dir = os.path.join(results_dir, FLAGS.results_dir)
temp_final_fp = os.path.join(results_dir, temp_fp)
while os.path.exists(temp_final_fp):
counter += 1
temp_fp = str(counter) + '_' + result_file_name
temp_final_fp = os.path.join(results_dir, temp_fp)
result_file_name = temp_fp
sub_result_path = os.path.join(results_dir, result_file_name)
accuracies = whitebox(
gan,
rec_data_path=FLAGS.rec_path,
batch_size=FLAGS.batch_size,
learning_rate=FLAGS.learning_rate,
nb_epochs=FLAGS.nb_epochs,
eps=FLAGS.fgsm_eps,
online_training=FLAGS.online_training,
defense_type=FLAGS.defense_type,
num_tests=FLAGS.num_tests,
attack_type=FLAGS.attack_type,
num_train=FLAGS.num_train,
)
ensure_dir(results_dir)
with open(sub_result_path, 'a') as f:
f.writelines([str(accuracies[i]) + ' ' for i in range(2)])
f.write('\n')
print('[*] saved accuracy in {}'.format(sub_result_path))
if accuracies[2]: # For attack detection.
pkl_result_path = sub_result_path.replace('.txt', '_roc.pkl')
with open(pkl_result_path, 'w') as f:
cPickle.dump(accuracies[2], f, cPickle.HIGHEST_PROTOCOL)
print('[*] saved roc_info in {}'.format(pkl_result_path))
def blackbox(gan,
rec_data_path=None,
batch_size=128,
learning_rate=0.001,
nb_epochs=10,
holdout=150,
data_aug=6,
nb_epochs_s=10,
lmbda=0.1,
online_training=False,
train_on_recs=False,
test_on_dev=True,
defense_type='none'):
"""MNIST tutorial for the black-box attack from arxiv.org/abs/1602.02697
Args:
train_start: index of first training set example
train_end: index of last training set example
test_start: index of first test set example
test_end: index of last test set example
defense_type: Type of defense against blackbox attacks
Returns:
a dictionary with:
* black-box model accuracy on test set
* substitute model accuracy on test set
* black-box model accuracy on adversarial examples transferred
from the substitute model
"""
FLAGS = flags.FLAGS
# Set logging level to see debug information.
set_log_level(logging.WARNING)
# Dictionary used to keep track and return key accuracies.
accuracies = {}
# Create TF session.
adv_training = False
if defense_type:
if defense_type == 'defense_gan' and gan:
sess = gan.sess
gan_defense_flag = True
else:
gan_defense_flag = False
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
if 'adv_tr' in defense_type:
adv_training = True
else:
gan_defense_flag = False
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
train_images, train_labels, test_images, test_labels = \
get_cached_gan_data(gan, test_on_dev, orig_data_flag=True)
x_shape, classes = list(train_images.shape[1:]), train_labels.shape[1]
nb_classes = classes
type_to_models = {
'A': model_a,
'B': model_b,
'C': model_c,
'D': model_d,
'E': model_e,
'F': model_f,
'Q': model_q,
'Z': model_z
}
bb_model = type_to_models[FLAGS.bb_model](
input_shape=[None] + x_shape,
nb_classes=train_labels.shape[1],
)
sub_model = type_to_models[FLAGS.sub_model](
input_shape=[None] + x_shape,
nb_classes=train_labels.shape[1],
)
if FLAGS.debug:
train_images = train_images[:20 * batch_size]
train_labels = train_labels[:20 * batch_size]
debug_dir = os.path.join('debug', 'blackbox', FLAGS.debug_dir)
ensure_dir(debug_dir)
x_debug_test = test_images[:batch_size]
# Initialize substitute training set reserved for adversary
images_sub = test_images[:holdout]
labels_sub = np.argmax(test_labels[:holdout], axis=1)
# Redefine test set as remaining samples unavailable to adversaries
if FLAGS.num_tests > 0:
test_images = test_images[:FLAGS.num_tests]
test_labels = test_labels[:FLAGS.num_tests]
test_images = test_images[holdout:]
test_labels = test_labels[holdout:]
# Define input and output TF placeholders
if FLAGS.image_dim[0] == 3:
FLAGS.image_dim = [
FLAGS.image_dim[1], FLAGS.image_dim[2], FLAGS.image_dim[0]
]
images_tensor = tf.placeholder(tf.float32, shape=[None] + x_shape)
labels_tensor = tf.placeholder(tf.float32, shape=(None, classes))
rng = np.random.RandomState([11, 24, 1990])
tf.set_random_seed(11241990)
train_images_bb, train_labels_bb, test_images_bb, test_labels_bb = \
train_images, train_labels, test_images, \
test_labels
cur_gan = None
if defense_type:
if 'gan' in defense_type:
# Load cached dataset reconstructions.
if online_training and not train_on_recs:
cur_gan = gan
elif not online_training and rec_data_path:
train_images_bb, train_labels_bb, test_images_bb, \
test_labels_bb = get_cached_gan_data(
gan, test_on_dev, orig_data_flag=False)
else:
assert not train_on_recs
if FLAGS.debug:
train_images_bb = train_images_bb[:20 * batch_size]
train_labels_bb = train_labels_bb[:20 * batch_size]
# Prepare the black_box model.
prep_bbox_out = prep_bbox(sess,
images_tensor,
labels_tensor,
train_images_bb,
train_labels_bb,
test_images_bb,
test_labels_bb,
nb_epochs,
batch_size,
learning_rate,
rng=rng,
gan=cur_gan,
adv_training=adv_training,
cnn_arch=bb_model)
else:
prep_bbox_out = prep_bbox(sess,
images_tensor,
labels_tensor,
train_images_bb,
train_labels_bb,
test_images_bb,
test_labels_bb,
nb_epochs,
batch_size,
learning_rate,
rng=rng,
gan=cur_gan,
adv_training=adv_training,
cnn_arch=bb_model)
model, bbox_preds, accuracies['bbox'] = prep_bbox_out
# Train substitute using method from https://arxiv.org/abs/1602.02697
print("Training the substitute model.")
reconstructed_tensors = tf.stop_gradient(
gan.reconstruct(images_tensor,
batch_size=batch_size,
reconstructor_id=1))
model_sub, preds_sub = train_sub(
sess,
images_tensor,
labels_tensor,
model(reconstructed_tensors),
images_sub,
labels_sub,
nb_classes,
nb_epochs_s,
batch_size,
learning_rate,
data_aug,
lmbda,
rng=rng,
substitute_model=sub_model,
)
accuracies['sub'] = 0
# Initialize the Fast Gradient Sign Method (FGSM) attack object.
fgsm_par = {
'eps': FLAGS.fgsm_eps,
'ord': np.inf,
'clip_min': 0.,
'clip_max': 1.
}
if gan:
if gan.dataset_name == 'celeba':
fgsm_par['clip_min'] = -1.0
fgsm = FastGradientMethod(model_sub, sess=sess)
# Craft adversarial examples using the substitute.
eval_params = {'batch_size': batch_size}
x_adv_sub = fgsm.generate(images_tensor, **fgsm_par)
if FLAGS.debug and gan is not None: # To see some qualitative results.
reconstructed_tensors = gan.reconstruct(x_adv_sub,
batch_size=batch_size,
reconstructor_id=2)
x_rec_orig = gan.reconstruct(images_tensor,
batch_size=batch_size,
reconstructor_id=3)
x_adv_sub_val = sess.run(x_adv_sub,
feed_dict={
images_tensor: x_debug_test,
K.learning_phase(): 0
})
sess.run(tf.local_variables_initializer())
x_rec_debug_val, x_rec_orig_val = sess.run(
[reconstructed_tensors, x_rec_orig],
feed_dict={
images_tensor: x_debug_test,
K.learning_phase(): 0
})
save_images_files(x_adv_sub_val, output_dir=debug_dir, postfix='adv')
postfix = 'gen_rec'
save_images_files(x_rec_debug_val,
output_dir=debug_dir,
postfix=postfix)
save_images_files(x_debug_test, output_dir=debug_dir, postfix='orig')
save_images_files(x_rec_orig_val,
output_dir=debug_dir,
postfix='orig_rec')
return
if gan_defense_flag:
reconstructed_tensors = gan.reconstruct(
x_adv_sub,
batch_size=batch_size,
reconstructor_id=4,
)
num_dims = len(images_tensor.get_shape())
avg_inds = list(range(1, num_dims))
diff_op = tf.reduce_mean(tf.square(x_adv_sub - reconstructed_tensors),
axis=avg_inds)
outs = model_eval_gan(sess,
images_tensor,
labels_tensor,
predictions=model(reconstructed_tensors),
test_images=test_images,
test_labels=test_labels,
args=eval_params,
diff_op=diff_op,
feed={K.learning_phase(): 0})
accuracies['bbox_on_sub_adv_ex'] = outs[0]
accuracies['roc_info'] = outs[1]
print('Test accuracy of oracle on adversarial examples generated '
'using the substitute: ' + str(outs[0]))
else:
accuracy = model_eval(sess,
images_tensor,
labels_tensor,
model(x_adv_sub),
test_images,
test_labels,
args=eval_params,
feed={K.learning_phase(): 0})
print('Test accuracy of oracle on adversarial examples generated '
'using the substitute: ' + str(accuracy))
accuracies['bbox_on_sub_adv_ex'] = accuracy
return accuracies
if __name__ == '__main__':
disable_all_warnings()
parser = get_common_parser()
parser.add_argument('--shape2', required=True, help='The shape to compare against', choices=allowed_shape_names())
parser.add_argument('--scaling2', help='The scaling used when fitting shape2. Default: none', choices=allowed_scaler_names())
parser.add_argument('--sigma_prior2', help='Prior to use for 1/sigma when fitting shape2. Default: None', choices=get_allowed_priors(is_sigma=True))
parser.add_argument('--priors2', help='The priors used for theta when fitting shape2. Default: None', choices=get_allowed_priors())
parser.add_argument('--filename', help='Where to save the figure. Default: results/comparison.png')
parser.add_argument('--show', help='Show figure and wait before exiting', action='store_true')
parser.add_argument('--ndiffs', type=int, default=5, help='Number of top diffs to show. Default=5.')
args = parser.parse_args()
data1, fitter1 = process_common_inputs(args)
data2 = get_data_from_args(args.dataset, args.pathway, args.from_age, args.scaling2, args.shuffle)
fitter2 = get_fitter_from_args(args.shape2, args.priors2, args.sigma_prior2)
fits1 = get_all_fits(data1,fitter1)
fits2 = get_all_fits(data2,fitter2)
print_diff_points(data1,fitter1,fits1, data2,fitter2,fits2, args.ndiffs)
fig = plot_comparison_scatter(data1,fitter1,fits1, data2,fitter2,fits2)
filename = args.filename
if filename is None:
ensure_dir(results_dir())
filename = join(results_dir(), 'shape_comparison.png')
save_figure(fig, filename)
if args.show:
plt.show(block=True)
def reconstruct_dataset(self, ckpt_path=None, max_num=-1, max_num_load=-1):
"""Reconstructs the images of the config's dataset with the generator.
"""
if not self.initialized:
self.load_generator(ckpt_path=ckpt_path)
splits = ['train', 'dev', 'test']
rec = self.reconstruct(self.real_data_test)
self.sess.run(tf.local_variables_initializer())
rets = {}
for split in splits:
if max_num > 0:
output_dir = os.path.join(
self.checkpoint_dir, 'recs_rr{:d}_lr{:.5f}_'
'iters{:d}_num{:d}'.format(self.rec_rr, self.rec_lr,
self.rec_iters, max_num), split)
else:
output_dir = os.path.join(
self.checkpoint_dir, 'recs_rr{:d}_lr{:.5f}_'
'iters{:d}'.format(self.rec_rr, self.rec_lr,
self.rec_iters), split)
if self.debug:
output_dir += '_debug'
ensure_dir(output_dir)
feats_path = os.path.join(output_dir, 'feats.pkl'.format(split))
could_load = False
try:
if os.path.exists(feats_path) and not self.test_again:
with open(feats_path) as f:
all_recs = cPickle.load(f)
could_load = True
print('[#] Successfully loaded features.')
else:
all_recs = []
except Exception as e:
all_recs = []
print('[#] Exception loading features {}'.format(str(e)))
gen_func = getattr(self, '{}_gen_test'.format(split))
all_targets = []
orig_imgs = []
ctr = 0
sti = time.time()
# Pickle files per reconstructed image.
pickle_out_dir = os.path.join(output_dir, 'pickles')
ensure_dir(pickle_out_dir)
single_feat_path_template = os.path.join(pickle_out_dir,
'rec_{:07d}_l{}.pkl')
for images, targets in gen_func():
batch_size = len(images)
im_paths = [
single_feat_path_template.format(ctr * batch_size + i,
targets[i])
for i in range(batch_size)
]
mn = max(max_num, max_num_load)
if (mn > -1 and ctr *
(len(images)) > mn) or (self.debug and ctr > 2):
break
batch_could_load = not self.test_again
batch_rec_list = []
for imp in im_paths: # Load per image cached files.
try:
with open(imp) as f:
loaded_rec = cPickle.load(f)
batch_rec_list.append(loaded_rec)
# print('[-] Loaded batch {}'.format(ctr))
except:
batch_could_load = False
break
if batch_could_load and not could_load:
recs = np.stack(batch_rec_list)
all_recs.append(recs)
if not (could_load or batch_could_load):
self.sess.run(tf.local_variables_initializer())
recs = self.sess.run(
rec,
feed_dict={self.real_data_test_pl: images},
)
print('[#] t:{:.2f} batch: {:d} '.format(
time.time() - sti, ctr))
all_recs.append(recs)
else:
print('[*] could load batch: {:d}'.format(ctr))
if not batch_could_load and not could_load:
for i in range(len(recs)):
pkl_path = im_paths[i]
with open(pkl_path, 'w') as f:
cPickle.dump(recs[i],
f,
protocol=cPickle.HIGHEST_PROTOCOL)
#print('[*] Saved reconstruction for {}'.format(pkl_path))
all_targets.append(targets)
orig_transformed = self.sess.run(
self.real_data_test,
feed_dict={self.real_data_test_pl: images})
orig_imgs.append(orig_transformed)
ctr += 1
if not could_load:
all_recs = np.concatenate(all_recs)
all_recs = all_recs.reshape([-1] + self.image_dim)
orig_imgs = np.concatenate(orig_imgs).reshape([-1] +
self.image_dim)
all_targets = np.concatenate(all_targets)
if self.debug:
save_images_files(all_recs,
output_dir=output_dir,
labels=all_targets)
save_images_files(
(orig_imgs + min(0, orig_imgs.min()) /
(orig_imgs.max() - min(0, orig_imgs.min()))),
output_dir=output_dir,
labels=all_targets,
postfix='_orig')
rets[split] = [all_recs, all_targets, orig_imgs]
return rets
