def make_private_params(self, run):
"""Initializes many experiment specific variables.
Parameters
----------
run : int
Number of the current experiment.
Returns
-------
private_params : dict
Directory that contains all experiment specific settings.
"""
private_params = {
'run':
run,
'seed':
self.params['random_seed'][run - 1],
'device':
'/gpu:*' if self.params['use_gpu'] else '/cpu:*',
'start_era':
0,
'current_era':
0,
'current_episode':
0,
'current_step':
0,
'eval_episodes':
0,
'eval_steps':
0,
'cumulative_gamma':
math.pow(self.params['gamma'], self.params['update_horizon'])
}
private_paths = {}
private_paths['base_dir'] = str(
Path(
os.path.join(
self.params['exp_dir'],
'run_{:02d}_s{}'.format(run, private_params['seed']))))
private_paths['ckpt_dir'] = str(
Path(os.path.join(private_paths['base_dir'], 'checkpoints')))
private_paths['log_dir'] = str(
Path(os.path.join(private_paths['base_dir'], 'logs')))
private_paths['var_file'] = str(
Path(os.path.join(private_paths['base_dir'], 'vars.yaml')))
if self.params['source_dir'] is not None:
util.copy_dir(self.params['source_dir'], private_paths['base_dir']) # pylint: disable=no-member
return private_params, private_paths
with open(tree_file, 'w') as f:
print( tree, file=f )
# reestimate model params for the given tree
cur_modelfile = raxml_launcher.evaluate(tree_file, ref_msa, cur_outdir)
# place outgroup
placement.launch_epa( tree_file, cur_modelfile, ref_msa, query_msa, cur_outdir, thorough=True )
result_files.append( os.path.join( cur_outdir, "epa_result.jplace") )
# ================================================================
# post-analysis
# ================================================================
result_dir = paths.epa_rooting_dir
util.make_path_clean( result_dir )
# do the summary high level stats
placement.outgroup_check( result_files, result_dir )
# also generate the lwr histograms
hist_csv_file = placement.gappa_examine_lwr( epa_out_dir, result_dir )
placement.ggplot_lwr_histogram( hist_csv_file, result_dir)
# ================================================================
# finally, export the results
# ================================================================
# also export the individual results
for f in result_files:
d = os.path.dirname( f )
util.copy_dir( d, os.path.join( result_dir, os.path.basename(d) ), ["*.rba", "*.phy", "*.startTree"] )
def test_copy_dir():
util.copy_dir('d1', 'd3', True)
return 'copy OK:'
def test_delete():
util.copy_file('a.txt', 'a.txt.bak')
util.copy_dir('d1', 'd1_bak')
util.delete('a.txt')
util.delete('d1', force=True)
return 'delete OK'
def main():
torch.multiprocessing.set_sharing_strategy('file_system')
print('[RUN] parse arguments')
args, framework, optimizer, data_loader_dict, tester_dict = option.parse_options()
print('[RUN] create result directories')
result_dir_dict = util.create_result_dir(args.result_dir, ['src', 'log', 'snapshot', 'test'])
util.copy_file(args.bash_file, args.result_dir)
util.copy_dir('./src', result_dir_dict['src'])
print('[RUN] create loggers')
train_log_dir = os.path.join(result_dir_dict['log'], 'train')
train_logger = SummaryWriter(train_log_dir)
print('[OPTIMIZER] learning rate:', optimizer.param_groups[0]['lr'])
n_batches = data_loader_dict['train'].__len__()
global_step = args.training_args['init_iter']
print('')
skip_flag = False
while True:
start_time = time.time()
for train_data_dict in data_loader_dict['train']:
batch_time = time.time() - start_time
if skip_flag:
skip_flag = False
else:
if global_step in args.snapshot_iters:
snapshot_dir = os.path.join(result_dir_dict['snapshot'], '%07d' % global_step)
util.save_snapshot(framework.network, optimizer, snapshot_dir)
if global_step in args.test_iters:
test_dir = os.path.join(result_dir_dict['test'], '%07d' % global_step)
util.run_testers(tester_dict, framework, data_loader_dict['test'], test_dir)
if args.training_args['max_iter'] <= global_step:
break
if global_step in args.training_args['lr_decay_schd'].keys():
util.update_learning_rate(optimizer, args.training_args['lr_decay_schd'][global_step])
train_loss_dict, train_time = \
train_network_one_step(args, framework, optimizer, train_data_dict, global_step)
if train_loss_dict is None:
skip_flag = True
train_data_dict.clear()
del train_data_dict
else:
if global_step % args.training_args['print_intv'] == 0:
iter_str = '[TRAINING] %d/%d:' % (global_step, args.training_args['max_iter'])
info_str = 'n_batches: %d, batch_time: %0.3f, train_time: %0.3f' % \
(n_batches, batch_time, train_time)
train_str = util.cvt_dict2str(train_loss_dict)
print(iter_str + '\n- ' + info_str + '\n- ' + train_str + '\n')
for key, value in train_loss_dict.items():
train_logger.add_scalar(key, value, global_step)
train_loss_dict.clear()
train_data_dict.clear()
del train_loss_dict, train_data_dict
global_step += 1
start_time = time.time()
if args.training_args['max_iter'] <= global_step:
break
train_logger.close()