trainer = Trainer(args, args_subset, args_dict_update)
trainer.sample(exp_num)
## SAMPLE
## -----------------------------
## Sample Prep.
del trainer
gc.collect()
print('Loading the best model and running the sample loop')
args_dict_update = {
'render': args.render,
'window_hop': 0,
'sample_all_styles': 0
}
## Sample
trainer = Trainer(args, args_subset, args_dict_update)
trainer.sample(exp_num)
## Finish
trainer.finish_exp()
## Print Experiment No.
print('\nExperiment Number: {}'.format(args.exp))
if __name__ == '__main__':
argparseNloop(loop)
## Sample
print('Loading the best model and running the sample loop')
args.__dict__.update({
'load':
book.name(book.weights_ext[0], book.weights_ext[1], args.save_dir)
})
sample(args, exp_num, data)
## Render (on a cpu only node)
# feats_kind_dict = {'rifke':'fke'}
# print('Rendering')
# render = Slurm('render', slurm_kwargs={'partition':'cpu_long', 'time':'10-00:00', 'n':10})
# python_cmd = ['source activate torch',
# 'python render.py -dataset {} -load {} -feats_kind {} -render_list {}'.format(
# args.dataset,
# args.load,
# feats_kind_dict[args.feats_kind],
# args.render_list)]
# render.run('\n'.join(python_cmd))
## Render new sentences
print('Rendering New Sentences')
render_new_sentences(args, exp_num, data)
# End Log
book._stop_log()
if __name__ == '__main__':
argparseNloop(train)
kwargs_dict.update({'h1':'{}'.format(path2videos)})
temp_filename = next(tempfile._get_candidate_names())
get_html_snippet('grid.html', 'templates/{}.html'.format(temp_filename), kwargs_dict)
file_list.append(temp_filename)
kwargs_dict = {'filenames':['{}.html'.format(file) for file in file_list]}
get_html_snippet('index.html', '{}.html'.format(temp_filename), kwargs_dict)
temp_srcs = ['htmlUtils/app/templates/{}.html'.format(file) for file in file_list]
src = 'htmlUtils/app/{}.html'.format(temp_filename)
dest = os.path.join(path2videos, '{}.html'.format(outfile))
shutil.move(src, dest)
for temp_src in temp_srcs:
os.remove(temp_src)
def makeHTMLfile_loop(args, exp_num):
assert args.load, 'Load file must be provided'
assert os.path.exists(args.load), 'Load file must exist'
args_subset = ['exp', 'cpk', 'speaker', 'model']
book = BookKeeper(args, args_subset, args_dict_update={'render':args.render},
tensorboard=args.tb)
args = book.args
dir_name = book.name.dir(args.save_dir)
makeHTMLfile(dir_name, idxs=args.render, outfile='videos')
makeHTMLfile(dir_name, idxs=4, outfile='videos_subset')
if __name__ == '__main__':
argparseNloop(makeHTMLfile_loop)
num_iter=500)
loss = criterion(y_cap, y)
for i_loss in internal_losses:
loss += i_loss
running_loss += loss.item() * batch_size
running_count += batch_size
if count >= 0 and args.debug: ## debugging by overfitting
break
return running_loss / running_count
train_loss = loop(model, data, train, pre, 'train')
dev_loss = loop(model, data, dev, pre, 'dev')
test_loss = loop(model, data, test, pre, 'test')
## update results but not save them
book.update_res({'train': train_loss, 'dev': dev_loss, 'test': test_loss})
## print results
book.print_res(0,
key_order=['train', 'dev', 'test'],
exp=exp_num,
lr=optim.param_groups[0]['lr'])
if __name__ == '__main__':
argparseNloop(sample)
save_animation(y_animates,
intervals,
dir_name,
desc,
data,
start,
subname=subname1,
text=texts,
output_modalities=output_modality,
mask=mask)
save_animation(y_animates_eval,
intervals,
dir_name,
desc,
data,
start,
subname=subname2,
text=texts,
output_modalities=output_modality,
mask=mask)
## render html
if set(keypoints_dirnames) - {'keypoints', 'keypoints_style'}:
makeHTMLfile(dir_name, args.render, 'videos')
makeHTMLfile(dir_name, 4, 'videos_subset')
if __name__ == '__main__':
argparseNloop(render)