nonlinear-axiom/plot_heatmap.py at main · watml/nonlinear-axiom · 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
import os
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

path_fig = os.path.join(
    'figs_heatmap',
    'dataset_id={}-heatmap.pdf'
    )

labels = ['1.5-norm', '2.5-norm', '5-norm', '10-norm', r'$\infty$-norm', 'ELC',
          'Banzhaf']
for key in [(16,1), (8,1), (4,1), (2,1), (1,1), (1,2), (1,4), (1,8), (1,16), (1,32)]:
    labels.append(str(key))


keys = [1.5, 2.5, 5, 10, np.inf, 'ELC', 0.5]


def plot_heatmap(curves, dataset_id):
    path_components = path_fig.split(os.sep)
    os.makedirs(os.sep.join(path_components[:-1]), exist_ok=True)


    candidates = []
    for key in keys:
        curve = curves[key]
        if curve[0,0,0] == np.inf:
            candidates.append(curve[:, -1, :].sum(axis=1))
        else:
            candidates.append(curve[:, :-1, :].sum(axis=-1).mean(axis=-1))

    for key in [(16,1), (8,1), (4,1), (2,1), (1,1), (1,2), (1,4), (1,8), (1,16), (1,32)]:
        curve = curves[key]
        candidates.append(curve[:, -1, :].sum(axis=1))


    A = np.empty((17, 17), dtype=np.float64)
    for i, row in enumerate(candidates):
        for j, column in enumerate(candidates):
            A[i, j] = ((row > column).sum() - (row < column).sum()) / 100


    fig, ax = plt.subplots(figsize=(32, 24))

    sns.heatmap(
        A,
        annot=True,
        fmt='.2f',
        vmin=0, vmax=1,
        xticklabels=labels,
        yticklabels=labels,
        linewidths=0.5,
        linecolor='white',
        ax=ax,
        annot_kws={'size': 20, 'weight': 'bold'},
        cbar_kws={'pad': 0.01}
        )

    ax.axhline(y=6, color='blue', linewidth=5)
    ax.axvline(x=6, color='blue', linewidth=5)

    cbar = ax.collections[0].colorbar
    cbar.ax.tick_params(labelsize=20)
    ax.set_xticklabels(ax.get_xticklabels(), rotation=90, ha='right', fontsize=20)
    ax.set_yticklabels(ax.get_yticklabels(), rotation=0, fontsize=20)


    plt.savefig(path_fig.format(dataset_id), bbox_inches='tight')
    plt.close(fig)


if __name__ == '__main__':
    from main import arg_dict
    from args import process_arg_dict
    from collections import defaultdict

    args = process_arg_dict(arg_dict)

    args_dataset = defaultdict(list)
    for arg in args:
        args_dataset[arg['dataset_id']].append(arg)

    for dataset_id, args_2nd in args_dataset.items():
        print(dataset_id)
        data = np.load(args_2nd[0]['path_data'])
        n_features = len(data['attribution'])

        curves = defaultdict(lambda : np.full((100, 6, n_features+1), np.inf, dtype=np.float64))

        for arg in args_2nd:
            data = np.load(arg['path_data'])
            auc = data['auc']

            if arg['method'] == 'semivalue':
                curves[arg['weight']][arg['instance_id'], -1] = auc
            else:
                if arg['random_seed_sampling'] == 'exact':
                    curves[arg['p']][arg['instance_id'], -1] = auc
                else:
                    curves[arg['p']][arg['instance_id'], arg['random_seed_sampling']] = auc

        plot_heatmap(curves, dataset_id)