DiffTF/helphurt.py at main · sadjadeb/DiffTF · 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
import csv
import numpy as np
import eval.ranking as rk
import ml_metrics as metrics
import matplotlib.pyplot as plt
import eval.load_dblp_data as dblp
import eval.evaluator as evaluator
import argparse

# === Parse arguments ===
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', type=str, choices=['dblp', 'dota2'], required=True)
parser.add_argument('--model1', type=int, required=True)
parser.add_argument('--model2', type=int, required=True)
args = parser.parse_args()

# === Configuration ===
filter_zero = True
at_k_mx = 10
at_k_set = range(1, at_k_mx + 1)

# === Load data ===
if args.dataset == 'dblp':
    file_names = [
        'Predictions/Rad_output_DBLP.csv',
        'Predictions/Sapienza_output_DBLP.csv',
        'Predictions/GERF_output_DBLP.csvv',
        'Predictions/DBLP_512_45k.csv'
    ]
    dataset_path = 'dataset/dblp/dblp_preprocessed_dataset_V2.3.pkl'
    index_path = 'dataset/dblp/Train_Test_Indices_V2.3.pkl'
    result_name = "Metrics/HelpHurt_{}_{}_DBLP.csv"
else:
    file_names = [
        'Predictions/DOTA2_rad_output.csv',
        'Predictions/DOTA2_sapienza_output.csv',
        'Predictions/DOTA2_GERF_output.csv',
        'Predictions/Dota2_1024.csv'
    ]
    dataset_path = 'dataset/dblp_preprocessed_dataset_V2.3.pkl'
    index_path = 'dataset/Train_Test_Indices_V2.3.pkl'
    result_name = "Metrics/HelpHurt_{}_{}_Dota2.csv"

model1 = args.model1 - 1
model2 = args.model2 - 1
file_name1 = file_names[model1]
file_name2 = file_names[model2]

foldIDsampleID_strata_dict = dblp.get_foldIDsampleID_stata_dict(
    data=dblp.load_preprocessed_dataset(file_path=dataset_path),
    train_test_indices=dblp.load_train_test_indices(file_path=index_path),
    kfold=10
)

# === Load outputs ===
method_name1, pred_indices1, true_indices1, _, _, k_fold, _ = evaluator.load_output_file(file_name1, foldIDsampleID_strata_dict)
method_name2, pred_indices2, true_indices2, _, _, _, _ = evaluator.load_output_file(file_name2, foldIDsampleID_strata_dict)
fold_set = np.arange(1, k_fold + 1)

# === Metric holders ===
holder_ndcg = evaluator.init_eval_holder(at_k_set)
holder_map = evaluator.init_eval_holder(at_k_set)
holder_mrr = evaluator.init_eval_holder(at_k_set)
if args.dataset == 'dblp':
    holder_recall = evaluator.init_eval_holder(at_k_set)

# === Compute differences ===
for k in at_k_set:
    for fold in fold_set:
        truth1 = true_indices1[fold]
        pred1 = pred_indices1[fold]
        truth2 = true_indices2[fold]
        pred2 = pred_indices2[fold]

        print(f'{method_name1} vs {method_name2} | Fold {fold} @K={k}')

        if args.dataset == 'dblp':
            holder_recall[k].extend([
                evaluator.r_at_k([p1], [t1], k=k) - evaluator.r_at_k([p2], [t2], k=k)
                for p1, t1, p2, t2 in zip(pred1, truth1, pred2, truth2)
            ])

        holder_ndcg[k].extend([
            rk.ndcg_at([p1], [t1], k=k) - rk.ndcg_at([p2], [t2], k=k)
            for p1, t1, p2, t2 in zip(pred1, truth1, pred2, truth2)
        ])
        holder_map[k].extend([
            metrics.mapk([p1], [t1], k=k) - metrics.mapk([p2], [t2], k=k)
            for p1, t1, p2, t2 in zip(pred1, truth1, pred2, truth2)
        ])
        holder_mrr[k].extend([
            evaluator.mean_reciprocal_rank(evaluator.cal_relevance_score([p1], [t1], k=k)) -
            evaluator.mean_reciprocal_rank(evaluator.cal_relevance_score([p2], [t2], k=k))
            for p1, t1, p2, t2 in zip(pred1, truth1, pred2, truth2)
        ])

# === Save metrics ===
output_path = result_name.format(method_name1, method_name2)
with open(output_path, 'w') as f:
    writer = csv.writer(f)
    header = ['@K']
    if args.dataset == 'dblp':
        header.append('Recall')
    header += ['NDCG', 'MAP', 'MRR']
    writer.writerow(header)

    for i, k in enumerate(at_k_set):
        row = [k]
        if args.dataset == 'dblp':
            row.append(np.mean(holder_recall[k]))
        row.append(np.mean(holder_ndcg[k]))
        row.append(np.mean(holder_map[k]))
        row.append(np.mean(holder_mrr[k]))
        writer.writerow(row)

print("File saved:", output_path)

# === Help-Hurt Plot ===
plt.title('Help-Hurt Plot')
plt.xlabel('Sample #')
if args.dataset == 'dblp':
    diff = np.sort(np.mean(list(holder_recall.values()), axis=0))[::-1]
else:
    diff = np.sort(np.mean(list(holder_ndcg.values()), axis=0))[::-1]

if filter_zero:
    diff = diff[diff != 0]

plt.bar(range(len(diff)), diff, color='b', width=1)
plt.plot(range(len(diff)), np.zeros(len(diff)), color='peachpuff')
plt.xlim(0, len(diff))
plt.ylim(-1.3 * max(abs(diff)), 1.3 * max(abs(diff)))
plt.grid()
plt.show()