siim-isic-melanoma-classification/ensemble.py at master · amri369/siim-isic-melanoma-classification · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
from trainer.predict import Predict
import pandas as pd
import os
import argparse
from scipy.stats import rankdata
from sklearn.metrics import roc_auc_score

class Ensemble(object):
    def __init__(self, params):
        # get learners with their trained weights
        self.params = params

    def predict(self, df, image_dir, validate=False, is_gpu_available=True):
        # initialization
        params = self.params
        archs = params['archs']
        paths = params['paths']
        sizes = params['sizes']
        names = params['names']

        # loop over predictors
        predictions = {}
        for arch, path, size, name in zip(archs, paths, sizes, names):
            # get predictor
            print('________Prediction for_____', arch, 'under name', name)
            predict = Predict(arch=arch,
                              classifier_path=path,
                              features_extractor_path=None,
                              is_gpu_available=is_gpu_available)

            # predict without data augmentation
            all_probs = predict(df, image_dir, validate=validate, size=size)
            predictions[name] = all_probs

        del predict

        predictions = pd.DataFrame(predictions)

        # add image names and targets
        predictions['image_name'] = df['image_name']
        try:
            predictions['target'] = df['target']
        except:
            predictions['target'] = -1
        columns = ['image_name'] + names + ['target']
        predictions = predictions[columns]

        return predictions

    @staticmethod
    def assemble(predictions, names, method='voting', weights=None):
        print('------Method------', method)
        df = predictions.copy()
        if method == 'voting':
            if weights is None:
                df['prediction'] = df[names].mean(axis=1)
            else:
                s = 0. * df[names[0]]
                for name, weight in zip(names, weights):
                    print(name, weight)
                    s = s + weight * df[name]
                df['prediction'] = s / s.max()

        if method == 'rankdata':
            for name in names:
                df[name] = rankdata(df[name], method='average')
                if weights is None:
                    df['prediction'] = df[names].sum(axis=1)
                    df['prediction'] = df['prediction'] / df['prediction'].max()
                else:
                    s = 0. * df[names[0]]
                    for name, weight in zip(names, weights):
                        print(name, weight)
                        s = s + weight * df[name]
                        df[name] = df[name] / df[name].max()
                    df['prediction'] = s / s.max()

        return df

def main(parser):
    # hard coded params
    #params = {}
    #params['archs'] = [
    #    'efficientnet-b7', 'efficientnet-b7', 'efficientnet-b7',
    #    'iternet_extractor_classifier_data', 'iternet_extractor_classifier_data', 'iternet_extractor_classifier_data',
    #    'residual_attention', 'residual_attention', 'residual_attention'
    #]
    #params['paths'] = [
    #    'exp/best-models/efficientnet-b7-K1-pretrained-ISIC2019.pth',
    #    'exp/best-models/efficientnet-b7-K2-pretrained-ISIC2019.pth',
    #    'exp/best-models/efficientnet-b7-K3-pretrained-ISIC2019.pth',
    #    'exp/best-models/iternet-K-1.pth',
    #    'exp/best-models/iternet-K-2.pth',
    #    'exp/best-models/iternet-K-3.pth',
    #    'exp/best-models/residual_attention-K-1.pth',
    #    'exp/best-models/residual_attention-K-2.pth',
    #    'exp/best-models/residual_attention-K-3.pth',
    #]
    #params['sizes'] = [224, 224, 224,
    #                   256, 256,
    #                   448, 448, 448
    #                  ]
    #params['names'] = ['efficientnet-b7-K-1', 'efficientnet-b7-K-2', 'efficientnet-b7-K-3',
    #                   'iternet-K-1', 'iternet-K-2',
    #                   'residual_attention-K-1', 'residual_attention-K-2', 'residual_attention-K-3'
    #                  ]

    # only efficient-nets
    params = {}
    params['archs'] = ['efficientnet-b7', 'efficientnet-b7', 'efficientnet-b7',
                       'iternet_extractor_classifier_data', 'iternet_extractor_classifier_data',
                       'iternet_extractor_classifier_data',
                       'residual_attention', 'residual_attention', 'residual_attention'
                      ]
    params['paths'] = [
        'exp/best-models/efficientnet-b7-K1-pretrained-ISIC2019.pth',
        'exp/best-models/efficientnet-b7-K2-pretrained-ISIC2019.pth',
        'exp/best-models/efficientnet-b7-K3-pretrained-ISIC2019.pth',
        'exp/best-models/iternet-K-1.pth',
        'exp/best-models/iternet-K-2.pth',
        'exp/best-models/iternet-K-3.pth',
        'exp/best-models/residual_attention-K-1.pth',
        'exp/best-models/residual_attention-K-2.pth',
        'exp/best-models/residual_attention-K-3.pth'
    ]
    params['sizes'] = [224, 224, 224,
                       256, 256, 256,
                       448, 448, 448
                      ]
    params['names'] = ['efficientnet-b7-K-1', 'efficientnet-b7-K-2', 'efficientnet-b7-K-3',
                      'iternet-K-1', 'iternet-K-2', 'iternet-K-3',
                      'residual_attention-K-1', 'residual_attention-K-2', 'residual_attention-K-3']

    # read dataframe
    df = pd.read_csv(args.csv_path)
    if not args.validate:
        df['target'] = 0
    else:
        target = df['target'].tolist()

    # predict using the ensemble method
    ensemble = Ensemble(params)
    predictions = ensemble.predict(df, args.image_dir, validate=args.validate, is_gpu_available=args.use_gpu)
    df = Ensemble.assemble(predictions, params['names'], method=args.ensemble_method, weights=None)

    # save predictions
    df.to_csv(os.path.join(args.results_path, 'results.csv'), index=None)
    df['target'] = df['prediction']
    df = df[['image_name', 'target']]
    df.to_csv(os.path.join(args.results_path, 'sample_submission.csv'), index=None)

    if args.validate:
        roc = roc_auc_score(target, df['target'].tolist())
        print('Final ROC', roc)

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='Model Training')
    parser.add_argument('--csv_path', default="data/val_split.csv",
                        type=str, help='path of csv file')
    parser.add_argument('--results_path', default="mohamed/submission",
                        type=str, help='loss type')
    parser.add_argument('--image_dir', default='data/resized-jpeg/train/',
                        type=str, help='Images folder path')
    parser.add_argument('--ensemble_method', default='voting',
                        type=str, help='ensemble method')
    parser.add_argument('--validate', action='store_true',
                        help='if to validate target')
    parser.add_argument('--use_gpu', action='store_true',
                        help='use gpu')

    args = parser.parse_args()

    main(args)