EdgeDependentNodeLabel/evaluate_baseline.py at master · young917/EdgeDependentNodeLabel · 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
import pandas as pd
import numpy as np
import os
import math
import argparse
import random
import sys
import utils
from matplotlib import pyplot as plt
from collections import defaultdict

from sklearn.metrics import accuracy_score, precision_score, recall_score, confusion_matrix
from sklearn.metrics import f1_score
from scipy.spatial import distance

import preprocess.data_load as dl

def get_clf_eval(y_test, pred, avg='micro'):
    confusion = confusion_matrix(y_test, pred)
    accuracy = accuracy_score(y_test, pred)
    precision = precision_score(y_test, pred, average=avg)
    recall = recall_score(y_test, pred, average=avg)
    f1 = f1_score(y_test, pred, average=avg)
    print('Confusion Matrix')
    print(confusion)
    print('Accuracy:{}, Precision:{}, Recall:{}, F1:{}'.format(accuracy, precision, recall, f1))

    return confusion, accuracy, precision, recall, f1

def get_result(Y_test, pred, dirname, name):
    confusion, accuracy, precision_micro, recall_micro, f1_micro = get_clf_eval(Y_test, pred, avg='micro')
    confusion, accuracy, precision_macro, recall_macro, f1_macro = get_clf_eval(Y_test, pred, avg='macro')
    result_dict = {
               "precision_micro" : precision_micro,
               "precision_macro" : precision_macro,
               "recall_micro" : recall_micro,
               "recall_macro" : recall_macro,
               "f1_micro" : f1_micro,
               "f1_macro" : f1_macro
    }

    base_output = dirname + "{}_confusion.txt".format(name)
    output_dim = confusion.shape[1]
    with open(base_output, "w") as f:
        for r in range(output_dim):
            for c in range(output_dim):
                f.write(str(confusion[r][c]))
                if c == (output_dim - 1):
                    f.write("\n")
                else:
                    f.write("\t")
    return result_dict

# Main ===========================================================================================
args = utils.parse_args()
outputdir = "./nongraph_results/" + args.dataset_name + "_" + str(args.k) + "/"
if args.evaltype == "test":
    assert args.fix_seed
    outputdir += "test/" + str(args.seed) + "/"
else:
    outputdir += "valid/"
if os.path.isdir(outputdir) is False:
    os.makedirs(outputdir)

if args.fix_seed:
    random.seed(args.seed)
    np.random.seed(args.seed)

# Data -----------------------------------------------------------------------------
data = dl.Hypergraph(args, args.dataset_name)
train_data = data.get_data(0)
if args.evaltype == "test":
    target_data = data.get_data(2)
else:
    target_data = data.get_data(1)

### Baseline 0
y_answer = []
y_pred = []
node2ansrole = {}
node2predrole = {}
for h in target_data:
    for v, vpos in zip(data.hedge2node[h], data.hedge2nodepos[h]):
        y_answer.append(vpos)
        vpos_pred = np.random.randint(args.output_dim)
        y_pred.append(vpos_pred)
        if v not in node2predrole:
            node2predrole[v] = np.zeros(args.output_dim)
            node2ansrole[v] = np.zeros(args.output_dim)
        node2predrole[v][vpos_pred] += 1
        node2ansrole[v][vpos] += 1
baseline0_result = get_result(y_answer, y_pred, outputdir, "baselineU")
# Calculate JSD-Divergence
jsd_div_list = []
try:
    for v in node2predrole.keys():
        jsd_div = distance.jensenshannon(node2ansrole[v], node2predrole[v])
        jsd_div_list.append(jsd_div)
    with open(outputdir + "baselineU_jsd_div.txt", "w") as f:
        for jsd_div in jsd_div_list:
            f.write(str(jsd_div) + "\n")
    print("Average: %.4f" % (np.mean(jsd_div_list)) )
except:
    print("except on jsd-divergence")
    with open("EXCEPTION.txt", "+a") as f:
        f.write(outputdir + " error on jsd-divergence\n")

### Baseline 1
label_dist = defaultdict(int)
y_pred = []
node2predrole = {}
total = 0
for h in train_data:
    for vpos in data.hedge2nodepos[h]:
        label_dist[int(vpos)] += 1.0
        total += 1
for lab in label_dist.keys():
    label_dist[lab] = label_dist[lab] / total
for h in target_data:
    for v, vpos in zip(data.hedge2node[h], data.hedge2nodepos[h]):
        vpos_pred = np.random.choice(3, 1, p=[label_dist[0], label_dist[1], label_dist[2]])[0]
        y_pred.append(vpos_pred)
        if v not in node2predrole:
            node2predrole[v] = np.zeros(args.output_dim)
        node2predrole[v][vpos_pred] += 1
baseline1_result = get_result(y_answer, y_pred, outputdir, "baselineP")
# Calculate JSD-Divergence
jsd_div_list = []
try:
    for v in node2predrole.keys():
        jsd_div = distance.jensenshannon(node2ansrole[v], node2predrole[v])
        jsd_div_list.append(jsd_div)
    with open(outputdir + "baselineP_jsd_div.txt", "w") as f:
        for jsd_div in jsd_div_list:
            f.write(str(jsd_div) + "\n")
    print("Average: %.4f" % (np.mean(jsd_div_list)) )
except:
    print("except on jsd-divergence")
    with open("EXCEPTION.txt", "+a") as f:
        f.write(outputdir + " error on jsd-divergence\n")