def main(args):
# create results directory if necessary
if not os.path.isdir(args.results_dir):
os.mkdir(args.results_dir)
if args.k_start == -1:
start = 0
else:
start = args.k_start
if args.k_end == -1:
end = args.k
else:
end = args.k_end
all_test_auc = []
all_val_auc = []
all_test_acc = []
all_val_acc = []
folds = np.arange(start, end)
for i in folds:
seed_torch(args.seed)
train_dataset, val_dataset, test_dataset = dataset.return_splits(
from_id=False,
csv_path='{}/splits_{}.csv'.format(args.split_dir, i))
datasets = (train_dataset, val_dataset, test_dataset)
results, test_auc, val_auc, test_acc, val_acc = train(
datasets, i, args)
all_test_auc.append(test_auc)
all_val_auc.append(val_auc)
all_test_acc.append(test_acc)
all_val_acc.append(val_acc)
#write results to pkl
filename = os.path.join(args.results_dir,
'split_{}_results.pkl'.format(i))
save_pkl(filename, results)
final_df = pd.DataFrame({
'folds': folds,
'test_auc': all_test_auc,
'val_auc': all_val_auc,
'test_acc': all_test_acc,
'val_acc': all_val_acc
})
if len(folds) != args.k:
save_name = 'summary_partial_{}_{}.csv'.format(start, end)
else:
save_name = 'summary.csv'
final_df.to_csv(os.path.join(args.results_dir, save_name))
def main(args):
# create results directory if necessary
if not os.path.isdir(args.results_dir):
os.mkdir(args.results_dir)
if args.k_start == -1:
start = 0
else:
start = args.k_start
if args.k_end == -1:
end = args.k
else:
end = args.k_end
val_cindex = []
folds = np.arange(start, end)
for i in folds:
start = timer()
seed_torch(args.seed)
train_dataset, val_dataset = dataset.return_splits(
from_id=False,
csv_path='{}/splits_{}.csv'.format(args.split_dir, i))
print('training: {}, validation: {}'.format(len(train_dataset),
len(val_dataset)))
datasets = (train_dataset, val_dataset)
if 'omic' in args.mode:
args.omic_input_dim = train_dataset.genomic_features.shape[1]
print("Genomic Dimension", args.omic_input_dim)
val_df, cindex = train(datasets, i, args)
val_cindex.append(cindex)
#write results to pkl
save_pkl(
os.path.join(args.results_dir,
'split_train_{}_results.pkl'.format(i)), val_df)
end = timer()
print('Fold %d Time: %f seconds' % (i, end - start))
if len(folds) != args.k:
save_name = 'summary_partial_{}_{}.csv'.format(start, end)
else:
save_name = 'summary.csv'
results_df = pd.DataFrame({'folds': folds, 'val_cindex': val_cindex})
results_df.to_csv(os.path.join(args.results_dir, 'summary_latest.csv'))