from __future__ import print_function
import numpy as np
import argparse
import torch
import torch.nn as nn
import pdb
import os
import pandas as pd
from utils.utils import *
from math import floor
import matplotlib.pyplot as plt
from datasets.dataset_generic import Generic_WSI_Classification_Dataset, Generic_MIL_Dataset
#, save_splits
from datasets.dataset_mtl import Generic_MIL_MTL_Dataset
#, save_splits
import h5py
from utils.eval_utils import eval
from utils.eval_utils_mtl import eval as eval_mtl
# Training settings
parser = argparse.ArgumentParser(description='CLAM Evaluation Script')
parser.add_argument('--data_root_dir', type=str, default='/media/fedshyvana/ssd1',
help='data directory')
parser.add_argument('--results_dir', type=str, default='./results',
help='relative path to results folder, i.e. '+
'the directory containing models_exp_code relative to project root (default: ./results)')
parser.add_argument('--save_exp_code', type=str, default=None,
help='experiment code to save eval results')
parser.add_argument('--models_exp_code', type=str, default=None,
help='experiment code to load trained models (directory under results_dir containing model checkpoints')
parser.add_argument('--splits_dir', type=str, default=None,
help='splits directory, if using custom splits other than what matches the task (default: None)')
parser.add_argument('--model_size', type=str, choices=['small', 'big'], default='big',
help='size of model (default: big)')
parser.add_argument('--model_type', type=str, choices=['clam', 'mil', 'attention_mil', 'clam_simple','histogram_mil'], default='attention_mil',
help='type of model (default: attention_mil)')
parser.add_argument('--drop_out', action='store_true', default=False,
help='whether model uses dropout')
parser.add_argument('--calc_features', action='store_true', default=False,
help='calculate features for pca/tsne')
parser.add_argument('--k', type=int, default=1, help='number of folds (default: 10)')
parser.add_argument('--k_start', type=int, default=-1, help='start fold (default: -1, last fold)')
parser.add_argument('--k_end', type=int, default=-1, help='end fold (default: -1, first fold)')
parser.add_argument('--fold', type=int, default=-1, help='single fold to evaluate')
parser.add_argument('--micro_average', action='store_true', default=False,
help='use micro_average instead of macro_avearge for multiclass AUC')
parser.add_argument('--mtl', action='store_true', default=False, help='flag to enable multi-task problem')
parser.add_argument('--patient_level', action='store_true', default=False, help='To enable computing scores at the patient-level. I.e. all patients slides are treated as a single bag with a single label')
parser.add_argument('--split', type=str, choices=['train', 'val', 'test', 'all'], default='test')
parser.add_argument('--task', type=str,
choices=['cardiac-grade','cardiac-mtl'])
args = parser.parse_args()
device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
encoding_size = 1024
args.save_dir = os.path.join('./eval_results', 'EVAL_' + str(args.save_exp_code))
args.models_dir = os.path.join(args.results_dir, str(args.models_exp_code))
os.makedirs(args.save_dir, exist_ok=True)
os.makedirs(os.path.join(args.save_dir, 'attention_scores'), exist_ok=True)
if args.splits_dir is None:
args.splits_dir = args.models_dir
assert os.path.isdir(args.models_dir)
assert os.path.isdir(args.splits_dir)
settings = {'task': args.task,
'split': args.split,
'save_dir': args.save_dir,
'models_dir': args.models_dir,
'model_type': args.model_type,
'drop_out': args.drop_out,
'model_size': args.model_size,
'micro_average': args.micro_average}
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 == 'cardiac-grade':
args.n_classes=2
dataset = Generic_MIL_Dataset(csv_path = 'dataset_csv/CardiacDummy_Grade.csv',
data_dir= os.path.join(args.data_root_dir, 'features'),
shuffle = False,
print_info = True,
label_dict = {'low':0, 'high':1},
patient_strat= False,
ignore=[],
patient_level = args.patient_level)
elif args.task == 'cardiac-mtl':
args.n_classes = [2,2,2]
dataset = Generic_MIL_MTL_Dataset(csv_path = 'dataset_csv/CardiacDummy_MTL.csv',
data_dir= os.path.join(args.data_root_dir, 'features'),
shuffle = False,
print_info = True,
label_dicts = [{'no_cell':0, 'cell':1},
{'no_amr':0, 'amr':1},
{'no_quilty':0, 'quilty':1}],
label_cols=['label_cell','label_amr','label_quilty'],
patient_strat= False,
ignore=[],
patient_level = args.patient_level)
elif os.path.isdir(args.task):
print('reading directory for fast inference')
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
if args.fold == -1:
folds = range(start, end)
else:
folds = range(args.fold, args.fold+1)
ckpt_paths = [os.path.join(args.models_dir, 's_{}_checkpoint.pt'.format(fold)) for fold in folds]
datasets_id = {'train': 0, 'val': 1, 'test': 2, 'all': -1}
def main(args):
all_auc = []
all_acc = []
all_aucs = []
for ckpt_idx in range(len(ckpt_paths)):
if datasets_id[args.split] < 0:
split_dataset = dataset
else:
csv_path = '{}/splits_{}.csv'.format(args.splits_dir, folds[ckpt_idx])
datasets = dataset.return_splits(from_id=False, csv_path=csv_path)
split_dataset = datasets[datasets_id[args.split]]
model, patient_results, test_error, auc, aucs, df = eval(split_dataset, args, ckpt_paths[ckpt_idx])
all_auc.append(auc)
all_acc.append(1-test_error)
if len(aucs) > 0:
all_aucs.append(aucs)
df.to_csv(os.path.join(args.save_dir, 'fold_{}.csv'.format(folds[ckpt_idx])), index=False)
if args.calc_features:
compute_features(split_dataset, args, ckpt_paths[ckpt_idx], args.save_dir, model=model)
df_dict = {'folds': folds, 'test_auc': all_auc, 'test_acc': all_acc}
if args.n_classes > 2:
all_aucs = np.vstack(all_aucs)
for i in range(args.n_classes):
df_dict.update({'class_{}_ovr_auc'.format(i):all_aucs[:,i]})
final_df = pd.DataFrame(df_dict)
if len(folds) != args.k:
save_name = 'summary_partial_{}_{}.csv'.format(folds[0], folds[-1])
else:
save_name = 'summary.csv'
final_df.to_csv(os.path.join(args.save_dir, save_name))
def main_mtl(args):
all_task1_auc = []
all_task1_acc = []
all_task2_auc = []
all_task2_acc = []
all_task3_auc = []
all_task3_acc = []
for ckpt_idx in range(len(ckpt_paths)):
if datasets_id[args.split] < 0:
split_dataset = dataset
else:
csv_path = '{}/splits_{}.csv'.format(args.splits_dir, folds[ckpt_idx])
datasets = dataset.return_splits(from_id=False, csv_path=csv_path)
split_dataset = datasets[datasets_id[args.split]]
model, results_dict = eval_mtl(split_dataset, args, ckpt_paths[ckpt_idx])
all_task1_auc.append(results_dict['auc_task1'])
all_task1_acc.append(1-results_dict['test_error_task1'])
all_task2_auc.append(results_dict['auc_task2'])
all_task2_acc.append(1-results_dict['test_error_task2'])
all_task3_auc.append(results_dict['auc_task3'])
all_task3_acc.append(1-results_dict['test_error_task3'])
df = results_dict['df']
df.to_csv(os.path.join(args.save_dir, 'fold_{}.csv'.format(folds[ckpt_idx])), index=False)
if args.calc_features:
compute_features(split_dataset, args, ckpt_paths[ckpt_idx], args.save_dir, model=model)
df_dict = {'folds': folds,
'task1_test_auc': all_task1_auc, 'task1_test_acc': all_task1_acc,
'task2_test_auc': all_task2_auc, 'task2_test_acc': all_task2_acc,
'task3_test_auc': all_task3_auc, 'task3_test_acc': all_task3_acc}
final_df = pd.DataFrame(df_dict)
if len(folds) != args.k:
save_name = 'summary_partial_{}_{}.csv'.format(folds[0], folds[-1])
else:
save_name = 'summary.csv'
final_df.to_csv(os.path.join(args.save_dir, save_name))
if __name__ == "__main__":
if args.mtl:
main_mtl(args)
else:
main(args)
print("finished!")
print("end script")
Datasets
Models
