print('Pearson Correlation between feature corr and blosum',
pearsonr(corr_scores, blos_scores)[0])
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('exp_name', help='Name of experiment')
args = parser.parse_args()
min_total = 100
filetype = 'pdf'
cache_path = get_cache_path() / f'{args.exp_name}.pickle'
args, feature_to_weighted_sums, weight_totals = pickle.load(
open(cache_path, "rb"))
print(args)
print(weight_totals)
report_dir = get_reports_path(
) / 'attention_analysis/blosum' / args.exp_name
report_dir.mkdir(parents=True, exist_ok=True)
create_figures(feature_to_weighted_sums, weight_totals, min_total,
report_dir, filetype)
with open(report_dir / 'args.json', 'w') as f:
json.dump(vars(args), f)
elif feature_name == 'ss4':
features.append(SecStructFeature())
elif feature_name == 'binding_sites':
features.append(BindingSiteFeature())
elif feature_name == 'protein_modifications':
features.append(ProteinModificationFeature())
elif feature_name == 'contact_map':
features.append(ContactMapFeature())
else:
raise ValueError(f"Invalid feature name: {feature_name}")
feature_to_weighted_sum, weight_total = compute_mean_attention(
model,
num_layers,
num_heads,
items,
features,
tokenizer,
args.model,
model_version,
cuda,
max_seq_len=args.max_seq_len,
min_attn=args.min_attn)
cache_dir = get_cache_path()
pathlib.Path(cache_dir).mkdir(parents=True, exist_ok=True)
path = cache_dir / f'{args.exp_name}.pickle'
pickle.dump((args, dict(feature_to_weighted_sum), weight_total),
open(path, 'wb'))
print('Wrote to', path)
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('exp_names', nargs='+', help='Names of experiments')
args = parser.parse_args()
min_total = 100
filetype = 'pdf'
report_dir = get_reports_path() / 'attention_analysis/edge_features_combined'
pathlib.Path(report_dir).mkdir(parents=True, exist_ok=True)
feature_data = []
include_features = [contact_map_pattern, secondary_to_pattern, binding_site_pattern]
for exp in args.exp_names:
cache_path = get_cache_path() / f'{exp}.pickle'
args, feature_to_weighted_sum, weight_total = pickle.load(open(cache_path, "rb"))
with open(report_dir / f'args_{exp}.json', 'w') as f:
json.dump(vars(args), f)
for feature, weighted_sum in feature_to_weighted_sum.items():
for p in include_features:
m = p.match(feature)
desc = None
if m:
if p == contact_map_pattern:
desc = 'Contact'
elif p == binding_site_pattern:
desc = 'Binding Site'
elif p == secondary_to_pattern:
sec = m.group(1)
desc = ss4_names.get(sec)