}
with open(args.save_dir + '/eval_experiment_{}.txt'.format(args.save_exp_code),
'w') as f:
print(settings, file=f)
f.close()
print(settings)
if args.task == 'task_1_tumor_vs_normal':
args.n_classes = 2
dataset = Generic_MIL_Dataset(
csv_path='dataset_csv/tumor_vs_normal_dummy_clean.csv',
data_dir=os.path.join(args.data_root_dir,
'tumor_vs_normal_resnet_features'),
shuffle=False,
print_info=True,
label_dict={
'normal_tissue': 0,
'tumor_tissue': 1
},
patient_strat=False,
ignore=[])
elif args.task == 'task_2_tumor_subtyping':
args.n_classes = 3
dataset = Generic_MIL_Dataset(
csv_path='dataset_csv/tumor_subtyping_dummy_clean.csv',
data_dir=os.path.join(args.data_root_dir,
'tumor_subtyping_resnet_features'),
shuffle=False,
print_info=True,
label_dict={
settings.update({
'bag_weight': args.bag_weight,
'inst_loss': args.inst_loss,
'B': args.B
})
print('\nLoad Dataset')
if args.task == 'camelyon_40x_cv':
args.n_classes = 2
dataset = Generic_MIL_Dataset(csv_path='dataset_csv/camelyon_clean.csv',
data_dir=os.path.join(
args.data_root_dir,
'camelyon_feat_resnet'),
shuffle=False,
seed=args.seed,
print_info=True,
label_dict={
'normal_tissue': 0,
'tumor_tissue': 1
},
patient_strat=False,
ignore=[])
elif args.task == 'tcga_kidney_cv':
args.n_classes = 3
dataset = Generic_MIL_Dataset(csv_path='dataset_csv/tcga_kidney_clean.csv',
data_dir=os.path.join(
args.data_root_dir,
'tcga_kidney_resnet_features'),
shuffle=False,
seed=args.seed,
'bag_weight': args.bag_weight,
'inst_loss': args.inst_loss,
'B': args.B
})
print('\nLoad Dataset')
if args.task == 'task_1_tumor_vs_normal':
args.n_classes = 2
dataset = Generic_MIL_Dataset(
csv_path='dataset_csv/tumor_vs_normal_dummy_clean.csv',
data_dir=os.path.join(args.data_root_dir,
'tumor_vs_normal_resnet_features'),
shuffle=False,
seed=args.seed,
print_info=True,
label_dict={
'normal_tissue': 0,
'tumor_tissue': 1
},
patient_strat=False,
ignore=[])
elif args.task == 'task_2_tumor_subtyping':
args.n_classes = 3
dataset = Generic_MIL_Dataset(
csv_path='dataset_csv/tumor_subtyping_dummy_clean.csv',
data_dir=os.path.join(args.data_root_dir,
'tumor_subtyping_resnet_features'),
shuffle=False,
seed=args.seed,
}
with open(args.save_dir + '/eval_experiment_{}.txt'.format(args.save_exp_code),
'w') as f:
print(settings, file=f)
f.close()
print(settings)
if args.task == 'camelyon_40x_cv':
args.n_classes = 2
dataset = Generic_MIL_Dataset(csv_path='dataset_csv/camelyon_clean.csv',
data_dir=os.path.join(
args.data_root_dir,
'camelyon_40x_resnet_features'),
shuffle=False,
print_info=True,
label_dict={
'normal_tissue': 0,
'tumor_tissue': 1
},
patient_strat=False,
ignore=[])
elif args.task == 'tcga_kidney_cv':
args.n_classes = 3
dataset = Generic_MIL_Dataset(csv_path='dataset_csv/tcga_kidney_clean.csv',
data_dir=os.path.join(
args.data_root_dir,
'tcga_kidney_resnet_features'),
shuffle=False,
print_info=True,
label_dict={
'model_size': args.model_size,
"use_drop_out": args.drop_out,
'weighted_sample': args.weighted_sample,
'opt': args.opt
}
print('\nLoad Dataset')
if args.task == 'camelyon_40x_cv':
args.n_classes = 2
dataset = Generic_MIL_Dataset(csv_path='dataset_csv/camelyon_clean.csv',
data_dir=os.path.join(
args.data_root_dir,
'camelyon_feat_resnet'),
shuffle=False,
seed=args.seed,
print_info=True,
label_dict={
'normal_tissue': 0,
'tumor_tissue': 1
},
patient_strat=False,
ignore=[])
elif args.task == 'colon_cancer':
args.n_classes = 2
dataset = Generic_MIL_Dataset(
csv_path='dataset_csv/label_information_colon.csv',
data_dir=os.path.join(args.data_root_dir, 'resnet_features_colon'),
shuffle=False,
seed=args.seed,
print_info=True,
'subtype_1': 0,
'subtype_2': 1,
'subtype_3': 2
},
patient_strat=True,
patient_voting='maj',
ignore=[])
elif args.task == 'hpa':
args.n_classes = 19
dataset = Generic_MIL_Dataset(csv_path='/content/label_df.csv',
data_dir='/content/feature',
shuffle=False,
seed=args.seed,
print_info=True,
label_dict={i: i
for i in range(19)},
patient_strat=False,
ignore=[],
hpa=True)
else:
raise NotImplementedError
num_slides_cls = np.array(
[len(cls_ids) for cls_ids in dataset.patient_cls_ids])
val_num = np.round(num_slides_cls * args.val_frac).astype(int)
test_num = np.round(num_slides_cls * args.test_frac).astype(int)
if __name__ == '__main__':
if args.label_frac > 0: