aes/eval_utils.py at master · sunnypwang/aes · 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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
import numpy as np
import pandas as pd
import os
from keras.callbacks import *
from sklearn.metrics import cohen_kappa_score
import utils
from data_utils import gen, prepare_features, rescale_to_int


class EvaluateCallback(Callback):
    def __init__(self, prompt, val_data, model_name, vocab=None, batch_size=5):
        self.prompt = prompt
        self.val_data = val_data
        self.model_name = model_name
        self.vocab = vocab
        self.batch_size = batch_size
        self.steps = np.ceil(len(val_data) / batch_size)
        self.y_true = prepare_features(model_name,
                                       df=val_data, prompt=prompt, y_only=True)

    def on_epoch_end(self, epoch, logs):
        y_pred = self.model.predict_generator(
            gen(self.model_name, self.prompt, self.val_data, self.vocab, self.batch_size, test=True, shuffle=False), steps=self.steps, verbose=1)

        generate_qwk(self.prompt, self.model_name,
                     self.y_true, y_pred, epoch+1, 'val')


def generate_qwk(prompt, model_name, y_true, y_pred, epoch, suffix=''):
    path = utils.mkpath('pred/{}'.format(model_name))

    y_true = rescale_to_int(y_true, prompt)
    y_pred = rescale_to_int(y_pred, prompt)
    qwk = QWK(y_true, y_pred)

    with open(os.path.join(path, 'qwk_{}_{}.csv'.format(prompt, suffix)), 'a+') as f:
        f.write('{}, {}\n'.format(epoch, qwk))


def generate_score(prompt, model_name, epoch, y_true, y_pred, aug_pred, test_df):
    path = utils.mkpath('pred/{}'.format(model_name))

    df = pd.DataFrame()
    df['essay_id'] = test_df['essay_id']
    df['essay_set'] = test_df['essay_set']
    df['domain1_score'] = y_true
    df['test'] = y_pred
    for key in aug_pred:
        df['test_' + key] = aug_pred[key]
    df.to_csv(os.path.join(path, 'score_{}_{}.tsv'.format(prompt, epoch)),
              sep='\t', index=False)
    return df


def generate_robustness(prompt, model_name, epoch, y_true, y_pred, aug_pred):
    path = utils.mkpath('pred/{}'.format(model_name))

    # y_true = rescale_to_int(y_true, prompt)
    y_pred_int = rescale_to_int(y_pred, prompt)
    aug_pred_int = {}
    wr_t, br_t, w_t, b_t = 0, 0, 0, 0
    N = len(y_pred) * len(aug_pred)
    print('N :', N)

    with open(os.path.join(path, 'robustness_{}_{}.csv'.format(prompt, epoch)), 'w+') as f:
        f.write('augment,worse_raw,better_raw,worse_resolved,better_resolved\n')
        for key in aug_pred:
            aug_pred_int[key] = rescale_to_int(aug_pred[key], prompt)

            wr, br, w, b = robustness(
                y_pred, aug_pred[key], y_pred_int, aug_pred_int[key])
            wr_t += wr
            br_t += br
            w_t += w
            b_t += b
            f.write('{},{},{},{},{}\n'.format(key, wr, br, w, b))
        f.write('sum,{},{},{},{}\n'.format(wr_t, br_t, w_t, b_t))
        f.write('avg,{},{},{},{}\n'.format(wr_t/N, br_t/N, w_t/N, b_t/N))


def generate_summary(model_name, epoch):
    prompts = [1, 2, 3, 4, 5, 6, 7, 8]
    # number of essay in test set
    length = [-1, 179, 180, 173, 177, 181, 180, 157, 73]
    path = utils.mkpath('pred/{}'.format(model_name))

    with open(os.path.join(path, 'summary_{}.txt'.format(epoch)), 'w+') as f:
        f.write('{} epoch {}\n\n'.format(model_name, epoch))
        f.write('QWK\n')
        qwk_avg = 0
        for p in prompts:
            qwk_df = pd.read_csv(os.path.join(path, 'qwk_{}_test.csv'.format(
                p)), header=None, names=['epoch', 'qwk'])
            qwk = qwk_df[qwk_df['epoch'] == epoch].values[-1, -1]
            f.write('{}\t{}\n'.format(p, qwk))
            qwk_avg += qwk

        f.write('\nRobustness per prompt\n')
        r_avg = 0
        r_aug_avg = 0
        for p in prompts:
            robustness_df = pd.read_csv(os.path.join(
                path, 'robustness_{}_{}.csv'.format(p, epoch)))
            r = (robustness_df['worse_resolved'] -
                 robustness_df['better_resolved']).values[-1]
            f.write('{}\t{}\n'.format(p, r))
            r_avg += r

            r_aug = (robustness_df['worse_resolved'] -
                     robustness_df['better_resolved']).values[:-2]/length[p]
            r_aug_avg += r_aug

        f.write('\nRobustness per augment\n')
        r_aug_avg /= 8
        for a, r in zip(robustness_df['augment'][:-2], r_aug_avg):
            f.write('{}\t{}\n'.format(a, r))

        f.write('\n')
        f.write('QWK Average:\t{}\n'.format(qwk_avg / 8))
        f.write('Robustness Average:\t{}\n'.format(r_avg / 8))
        f.write('Robustness Average:\t{}\n'.format(r_aug_avg.mean()))
    print('summary generated!')


def generate_summary_best(model_name):
    prompts = [1, 2, 3, 4, 5, 6, 7, 8]
    # number of essay in test set
    length = [-1, 179, 180, 173, 177, 181, 180, 157, 73]
    path = utils.mkpath('pred/{}'.format(model_name))

    best_ep = [-1]*9
    with open(os.path.join(path, 'summary_best.txt'), 'w+') as f:
        f.write('{}\n\n'.format(model_name))
        f.write('QWK\n')
        f.write('epoch\tprompt\tqwk\n')
        qwk_avg = 0
        for p in prompts:
            qwk_df = pd.read_csv(os.path.join(path, 'qwk_{}_val.csv'.format(
                p)), header=None, names=['epoch', 'qwk'])
            max_idx = qwk_df['qwk'].idxmax()
            best_ep[p] = int(qwk_df.iloc[max_idx].values[0])

            qwk_df = pd.read_csv(os.path.join(path, 'qwk_{}_test.csv'.format(
                p)), header=None, names=['epoch', 'qwk'])

            try:
                tmp = qwk_df[qwk_df['epoch'] == best_ep[p]].values
                # in case of multiple runs of same epoch, pick one with the best QWK
                ep, qwk = tmp[tmp.argmax(axis=0)[-1]]
            except:
                raise Exception(
                    'Error: epoch {} of prompt {} not found in test'.format(best_ep[p], p))

            f.write('{}\t{}\t{}\n'.format(best_ep[p], p, qwk))
            qwk_avg += qwk

        f.write('\nRobustness per prompt\n')
        r_avg = 0
        r_aug_avg = 0
        for p in prompts:
            robustness_df = pd.read_csv(os.path.join(
                path, 'robustness_{}_{}.csv'.format(p, best_ep[p])))
            r = (robustness_df['worse_resolved'] -
                 robustness_df['better_resolved']).values[-1]
            f.write('{}\t{}\n'.format(p, r))
            r_avg += r

            r_aug = (robustness_df['worse_resolved'] -
                     robustness_df['better_resolved']).values[:-2]/length[p]
            r_aug_avg += r_aug

        f.write('\nRobustness per augment\n')
        r_aug_avg /= 8
        for a, r in zip(robustness_df['augment'][:-2], r_aug_avg):
            f.write('{}\t{}\n'.format(a, r))

        f.write('\n')
        f.write('QWK Average:\t{}\n'.format(qwk_avg / 8))
        f.write('Robustness Average:\t{}\n'.format(r_avg / 8))
        f.write('Robustness Average:\t{}\n'.format(r_aug_avg.mean()))
    print('summary generated!')


def QWK(y_true, y_pred):
    return cohen_kappa_score(y_true, y_pred, weights='quadratic')


def robustness(original, augment, original_int, augment_int, threshold=0.0):
    worse_raw = np.sum(original - augment > threshold)
    better_raw = np.sum(augment - original > threshold)
    worse_resolved = np.sum(original_int > augment_int)
    better_resolved = np.sum(original_int < augment_int)
    return worse_raw, better_raw, worse_resolved, better_resolved