hyper_dim=int(model_weight[11]),
eps=eps))
n_outlier = len(outliers)
if n_outlier <= 50:
outlier_list.append(outliers)
break
np.save('outlier_list.npy', np.array(outlier_list))
outlier_list_uni, outlier_count = np.unique(np.hstack(outlier_list),
return_counts=True)
outlier_list_ulti = outlier_list_uni[np.where(outlier_count > 1)]
print('Writing pdb files')
# write the pdb according the outlier indices
traj_info = open('./scheduler_logs/openmm_log.txt', 'r').read().split()
traj_dict = dict(zip(traj_info[::2], np.array(traj_info[1::2]).astype(int)))
outliers_pdb = os.path.join(work_dir, 'outlier_pdbs')
make_dir_p(outliers_pdb)
for outlier in outlier_list_ulti:
traj_file, frame_number = find_frame(traj_dict, outlier)
outlier_pdb_file = os.path.join(
outliers_pdb, '%d_%s_%d.pdb' % (outlier, traj_file[:18], frame_number))
outlier_pdb = write_pdb_frame(traj_file, pdb_file, frame_number,
outlier_pdb_file)
print('Finishing and cleaning up the jobs. ')
subprocess.Popen('bash prerun_clean.sh'.split(" "))
if n_outlier <= 50:
outlier_list.append(outliers)
break
outlier_list_uni, outlier_count = np.unique(np.hstack(outlier_list), return_counts=True)
print('\nPreparing to write new pdb files')
# write the pdb according the outlier indices
traj_info = open('./scheduler_logs/openmm_log.txt', 'r').read().split()
traj_dict = dict(zip(traj_info[::2], np.array(traj_info[1::2]).astype(int)))
# Write the new outliers
n_outlier_iter = 0
new_outlier_list = []
for outlier in outlier_list_uni:
traj_file, num_frame = find_frame(traj_dict, outlier)
if num_frame == 0:
print('Detected initial point as outlier, skipping...')
continue
outlier_pdb_file = os.path.join(outliers_pdb_path, '{}_{:06d}.pdb'.format(traj_file[:18], num_frame))
new_outlier_list.append(outlier_pdb_file)
if outlier_pdb_file not in outlier_pdb_files:
print('Found a new outlier# {} at frame {} of {}'.format(outlier, num_frame, traj_file))
outlier_pdb = write_pdb_frame(traj_file, pdb_file, num_frame, outlier_pdb_file)
print(' Written as {}'.format(outlier_pdb_file))
outlier_pdb_files.append(outlier_pdb_file)
n_outlier_iter += 1
for outlier_pdb_file in outlier_pdb_files:
if outlier_pdb_file not in new_outlier_list:
print('Old outlier {} is now connected to a cluster and removing it from the outlier list '.format(outlier_pdb_file[-29:]))
outlier_list_ranked, _ = outliers_from_latent_loc(cm_predict,
n_outliers=n_outliers,
n_jobs=12)
print("Done outlier searching...")
# print(outlier_list_ranked)
# Write the outliers using MDAnalysis
outliers_pdb_path = os.path.abspath('./outlier_pdbs')
os.makedirs(outliers_pdb_path, exist_ok=True)
print('Writing outliers in %s' % outliers_pdb_path)
# identify new outliers
new_outliers_list = []
for outlier in outlier_list_ranked:
# find the location of outlier
traj_dir, num_frame = find_frame(traj_dict, outlier)
traj_file = os.path.join(traj_dir, 'output.dcd')
# get the outlier name - traj_label + frame number
run_name = os.path.basename(traj_dir)
pdb_name = f"{run_name}_{num_frame:06}.pdb"
outlier_pdb_file = os.path.join(outliers_pdb_path, pdb_name)
new_outliers_list.append(outlier_pdb_file)
# Only write new pdbs to reduce I/O redundancy.
if not os.path.exists(outlier_pdb_file):
print(f'New outlier at frame {num_frame} of {run_name}')
outlier_pdb = write_pdb_frame(traj_file, pdb_file, num_frame,
outlier_pdb_file)
# Clean up outdated outliers (just for bookkeeping)
outliers_list = glob(os.path.join(outliers_pdb_path, 'omm_runs*.pdb'))