-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgridsearch_evaluate.py
232 lines (204 loc) · 8.74 KB
/
gridsearch_evaluate.py
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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
import os
import re
import numpy as np
import pandas as pd
from pylatex import Document, Package, NoEscape
from utils.datasets import BASE_DIR
EVALUATION_DIR = f'{BASE_DIR}gs0/eval/'
BACKEND_NAMES = {
'btree': 'BT',
'vtree': 'LT',
'rtree': 'RT',
'ptree': 'RT-S',
'ctree': 'HCLT'
}
IGNORE = [
'device',
'seed',
'sam_valid',
'sam_unique',
'sam_unique_abs',
'sam_novel',
'sam_novel_abs',
'sam_score',
'sam_m_stab',
'sam_a_stab',
'sam_fcd_trn',
'sam_kldiv_trn',
'sam_nspdk_trn',
'sam_fcd_val',
'sam_kldiv_val',
'sam_nspdk_val',
'sam_fcd_tst',
'sam_kldiv_tst',
'sam_nspdk_tst',
'res_valid',
'res_unique',
'res_unique_abs',
'res_novel',
'res_novel_abs',
'res_score',
'res_m_stab',
'res_a_stab',
'res_fcd_trn',
'res_kldiv_trn',
'res_nspdk_trn',
'res_fcd_val',
'res_kldiv_val',
'res_nspdk_val',
'res_fcd_tst',
'res_kldiv_tst',
'res_nspdk_tst',
'cor_valid',
'cor_unique',
'cor_unique_abs',
'cor_novel',
'cor_novel_abs',
'cor_score',
'cor_m_stab',
'cor_a_stab',
'cor_fcd_trn',
'cor_kldiv_trn',
'cor_nspdk_trn',
'cor_fcd_val',
'cor_kldiv_val',
'cor_nspdk_val',
'cor_fcd_tst',
'cor_kldiv_tst',
'cor_nspdk_tst',
'nll_trn_approx',
'nll_val_approx',
'nll_tst_approx',
'time_sam',
'time_res',
'time_cor',
'atom_list',
'num_params',
'model_path'
]
COLUMN_NAMES = [
'Model', 'Valid', 'NSPDK', 'FCD', 'Unique', 'Novel']
def baseline_models_qm9():
rows = [
['GraphAF', r'74.43$\pm$2.55', r'0.021$\pm$0.003', r' 5.27$\pm$0.40', r'88.64$\pm$2.37', r'86.59$\pm$1.95'],
['GraphDF', r'93.88$\pm$4.76', r'0.064$\pm$0.000', r'10.93$\pm$0.04', r'98.58$\pm$0.25', r'98.54$\pm$0.48'],
['MoFlow', r'91.36$\pm$1.23', r'0.017$\pm$0.003', r' 4.47$\pm$0.60', r'98.65$\pm$0.57', r'94.72$\pm$0.77'],
['EDP-GNN', r'47.52$\pm$3.60', r'0.005$\pm$0.001', r' 2.68$\pm$0.22', r'99.25$\pm$0.05', r'86.58$\pm$1.85'],
['GraphEBM', r' 8.22$\pm$2.24', r'0.030$\pm$0.004', r' 6.14$\pm$0.41', r'97.90$\pm$0.14', r'97.01$\pm$0.17'],
['SPECTRE', r'87.30$\pm$n/a', r'0.163$\pm$n/a', r'47.96$\pm$n/a', r'35.70$\pm$n/a', r'97.28$\pm$n/a' ],
['GDSS', r'95.72$\pm$1.94', r'0.003$\pm$0.000', r' 2.90$\pm$0.28', r'98.46$\pm$0.61', r'86.27$\pm$2.29'],
['DiGress', r'99.00$\pm$0.10', r'0.005$\pm$n/a', r' 0.36$\pm$n/a', r'96.20$\pm$n/a', r'33.40$\pm$n/a' ],
['GRAPHARM', r'90.25$\pm$n/a', r'0.002$\pm$n/a', r' 1.22$\pm$n/a', r'95.62$\pm$n/a', r'70.39$\pm$n/a' ]
]
return pd.DataFrame(rows, columns=COLUMN_NAMES)
def baseline_models_zinc250k():
rows = [
['GraphAF', r'68.47$\pm$0.99', r'0.044$\pm$0.005', r'16.02$\pm$0.48', r'98.64$\pm$0.69', r'100.00$\pm$0.00'],
['GraphDF', r'90.61$\pm$4.30', r'0.177$\pm$0.001', r'33.55$\pm$0.16', r'99.63$\pm$0.01', r' 99.99$\pm$0.01'],
['MoFlow', r'63.11$\pm$5.17', r'0.046$\pm$0.002', r'20.93$\pm$0.18', r'99.99$\pm$0.01', r'100.00$\pm$0.00'],
['EDP-GNN', r'82.97$\pm$2.73', r'0.049$\pm$0.006', r'16.74$\pm$1.30', r'99.79$\pm$0.08', r'100.00$\pm$0.00'],
['GraphEBM', r' 5.29$\pm$3.83', r'0.212$\pm$0.005', r'35.47$\pm$5.33', r'98.79$\pm$0.15', r'100.00$\pm$0.00'],
['SPECTRE', r'90.20$\pm$n/a', r'0.109$\pm$n/a', r'18.44$\pm$n/a', r'67.05$\pm$n/a', r'100.00$\pm$n/a' ],
['GDSS', r'97.01$\pm$0.77', r'0.019$\pm$0.001', r'14.66$\pm$0.68', r'99.64$\pm$0.13', r'100.00$\pm$0.00'],
['DiGress', r'91.02$\pm$n/a', r'0.082$\pm$n/a', r'23.06$\pm$n/a', r'81.23$\pm$n/a', r'100.00$\pm$n/a' ],
['GRAPHARM', r'88.23$\pm$n/a', r'0.055$\pm$n/a', r'16.26$\pm$n/a', r'99.46$\pm$n/a', r'100.00$\pm$n/a' ]
]
return pd.DataFrame(rows, columns=COLUMN_NAMES)
def highlight_top3(x, type='max'):
x = [float(s.split('$\pm$')[0]) for s in x]
styles = np.array(len(x)*[None])
match type:
case 'min':
i = np.argsort(x)[:3]
case 'max':
i = np.argsort(x)[-3:][::-1]
case _:
os.error('Unsupported type.')
styles[i] = [
'color:{c1};textbf:--rwrap;',
'color:{c2};textbf:--rwrap;',
'color:{c3};textbf:--rwrap;'
]
return styles
def format_number(match):
number = float(match.group(1))
number_original = match.group(1)
if (number > 0.3 and number < 10) or (number > 99.98 and number < 100):
return r"\phantom{0}" + number_original
else:
return number_original
def latexify_style(df, path, row_names=None, column_names=None):
if row_names is not None:
df.replace(row_names, inplace=True)
if column_names is not None:
df.rename(columns={o: n for o, n in zip(df.columns, column_names)}, inplace=True)
subset_min = [('QM9','NSPDK'), ('QM9','FCD'), ('Zinc250k','NSPDK'), ('Zinc250k','FCD')]
subset_max = [('QM9','Valid'), ('QM9','Unique'), ('QM9','Novel'), ('Zinc250k','Valid'), ('Zinc250k','Unique'), ('Zinc250k','Novel')]
s = df.style
s = s.apply(highlight_top3, type='min', subset=subset_min)
s = s.apply(highlight_top3, type='max', subset=subset_max)
s.hide()
s.to_latex(
path,
# column_format='lcccccccccc',
hrules=True,
multicol_align='c'
)
line = []
with open(path, 'r') as file_data:
file_data = re.sub(r'(\d*\.?\d+)(?=\s*\$\\pm\$)', format_number, file_data.read())
lines = file_data.splitlines()
for w in lines[3].split():
match w:
case 'Valid' | 'Unique' | 'Novel':
w += r'$\uparrow$'
case 'NSPDK' | 'FCD':
w += r'$\downarrow$'
line.append(w)
lines[3] = ' '.join(line)
lines.insert(3, '\\cmidrule(lr){2-6}')
lines.insert(4, '\\cmidrule(lr){7-11}')
lines.insert(16, '\\midrule')
print('\n'.join(lines), file=open(path, 'w'))
def latexify_table(r_name, w_name, clean_tex=True):
with open(r_name) as f:
table = f.read()
doc = Document(documentclass='standalone', document_options=('preview'), geometry_options={'margin': '0cm'})
doc.packages.append(Package('booktabs'))
doc.packages.append(Package('xcolor', options='table'))
doc.packages.append(NoEscape(r'\definecolor{c1}{RGB}{27,158,119}'))
doc.packages.append(NoEscape(r'\definecolor{c2}{RGB}{117,112,179}'))
doc.packages.append(NoEscape(r'\definecolor{c3}{RGB}{217,95,2}'))
doc.append(NoEscape(table))
doc.generate_pdf(f'{w_name}', clean_tex=clean_tex)
def find_best(evaluation_dir, dataset, model, backends):
d_frame = pd.DataFrame('', index=range(len(backends.keys())), columns=COLUMN_NAMES)
path = evaluation_dir + f'metrics/{dataset}/{model}/'
for i, backend in enumerate(backends.keys()):
b_frame = pd.concat([pd.read_csv(path + f) for f in os.listdir(path) if backend in f])
g_frame = b_frame.groupby(list(filter(lambda x: x not in IGNORE, b_frame.columns)))
a_frame = g_frame.agg({'sam_fcd_val': 'mean'})
f_frame = g_frame.get_group(a_frame['sam_fcd_val'].idxmin())
d_frame.loc[i] = [
BACKEND_NAMES[backend],
f'{ 100*f_frame["sam_valid"].mean():.2f}$\\pm${ 100*f_frame["sam_valid"].std():.2f}',
f'{ f_frame["sam_nspdk_tst"].mean():.3f}$\\pm${ f_frame["sam_nspdk_tst"].std():.3f}',
f'{ f_frame["sam_fcd_trn"].mean():.2f}$\\pm${ f_frame["sam_fcd_trn"].std():.2f}',
f'{100*f_frame["sam_unique"].mean():.2f}$\\pm${100*f_frame["sam_unique"].std():.2f}',
f'{100*f_frame["sam_novel" ].mean():.2f}$\\pm${100*f_frame["sam_novel" ].std():.2f}'
]
return d_frame
if __name__ == "__main__":
baselines_qm9 = baseline_models_qm9()
ourmodels_qm9 = find_best(EVALUATION_DIR, 'qm9', 'marg_sort', BACKEND_NAMES)
allmodels_qm9 = pd.concat([baselines_qm9, ourmodels_qm9], ignore_index=True)
baselines_zinc250k = baseline_models_zinc250k()
ourmodels_zinc250k = find_best(EVALUATION_DIR, 'zinc250k', 'marg_sort', BACKEND_NAMES)
allmodels_zinc250k = pd.concat([baselines_zinc250k, ourmodels_zinc250k], ignore_index=True)
allmodels = allmodels_qm9.merge(allmodels_zinc250k, how='left', on='Model', suffixes=('-x', '-y'))
allmodels.columns = COLUMN_NAMES + COLUMN_NAMES[1:]
columns = [('','Model')] + [('QM9', name) for name in COLUMN_NAMES[1:]] + [('Zinc250k', name) for name in COLUMN_NAMES[1:]]
allmodels.columns = pd.MultiIndex.from_tuples(columns)
allmodels.head()
latexify_style(allmodels, 'results/unconditional.tab')
latexify_table('results/unconditional.tab', 'results/unconditional')