forked from microsoft/synthetic-data-showcase
-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluator.py
299 lines (262 loc) · 14.4 KB
/
evaluator.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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
import time
import datetime
import logging
from os import path
from collections import defaultdict
import util as util
def evaluate(config):
"""Evaluates the error in the synthetic microdata with respect to the sensitive microdata.
Produces output statistics (tsv) and graphics (svg) for preservation_by_length and preservation_by_count,
as well as leakage analysis from the synthetic data (as synthetic_leakage_by_length).
Args:
config: options from the json config file, else default values.
"""
use_columns = config['use_columns']
record_limit = config['record_limit']
reporting_length = config['reporting_length']
reporting_resolution = config['reporting_resolution']
sensitive_microdata_path = config['sensitive_microdata_path']
sensitive_microdata_delimiter = config['sensitive_microdata_delimiter']
synthetic_microdata_path = config['synthetic_microdata_path']
sensitive_zeros = config['sensitive_zeros']
parallel_jobs = config['parallel_jobs']
output_dir = config['output_dir']
prefix = config['prefix']
logging.info(f'Evaluate {synthetic_microdata_path} vs {sensitive_microdata_path}')
start_time = time.time()
sen_counts = None
sen_records = None
sen_df = util.loadMicrodata(
path=sensitive_microdata_path, delimiter=sensitive_microdata_delimiter, record_limit=record_limit,
use_columns=use_columns)
sen_records = util.genRowList(sen_df, sensitive_zeros)
if not path.exists(config['sensitive_aggregates_path']):
logging.info('Computing sensitive aggregates...')
if reporting_length == -1:
reporting_length = max([len(row) for row in sen_records])
sen_counts = util.countAllCombos(sen_records, reporting_length, parallel_jobs)
else:
logging.info('Loading sensitive aggregates...')
sen_counts = util.loadSavedAggregates(config['sensitive_aggregates_path'])
if reporting_length == -1:
reporting_length = max(sen_counts.keys())
if use_columns != []:
reporting_length = min(reporting_length, len(use_columns))
filtered_sen_counts = {length: {combo: count for combo, count in combo_to_counts.items(
) if count >= reporting_resolution} for length, combo_to_counts in sen_counts.items()}
syn_df = util.loadMicrodata(path=synthetic_microdata_path, delimiter='\t', record_limit=-1, use_columns=use_columns)
syn_records = util.genRowList(syn_df, sensitive_zeros)
syn_counts = util.countAllCombos(syn_records, reporting_length, parallel_jobs)
len_to_syn_count = {length: len(combo_to_count) for length, combo_to_count in syn_counts.items()}
len_to_sen_rare = {length: {combo: count for combo, count in combo_to_count.items() if count < reporting_resolution}
for length, combo_to_count in sen_counts.items()}
len_to_syn_rare = {length: {combo: count for combo, count in combo_to_count.items() if count < reporting_resolution}
for length, combo_to_count in syn_counts.items()}
len_to_syn_leak = {length: len([1 for rare in rares if rare in syn_counts.get(length, {}).keys()])
for length, rares in len_to_sen_rare.items()}
sen_unique_to_records, sen_rare_to_records, _ = util.mapShortestUniqueRareComboLengthToRecords(
sen_records, len_to_sen_rare)
sen_rare_to_sen_count = {length: util.protect(len(records), reporting_resolution)
for length, records in sen_rare_to_records.items()}
sen_unique_to_sen_count = {length: util.protect(len(records), reporting_resolution)
for length, records in sen_unique_to_records.items()}
total_sen = util.protect(len(sen_records), reporting_resolution)
unique_total = sum([v for k, v in sen_unique_to_sen_count.items() if k > 0])
rare_total = sum([v for k, v in sen_rare_to_sen_count.items() if k > 0])
risky_total = unique_total + rare_total
risky_total_pct = 100*risky_total/total_sen
record_analysis_tsv = path.join(output_dir, f'{prefix}_sensitive_analysis_by_length.tsv')
with open(record_analysis_tsv, 'w') as f:
f.write('\t'.join(['combo_length', 'sen_rare', 'sen_rare_pct',
'sen_unique', 'sen_unique_pct', 'sen_risky', 'sen_risky_pct'])+'\n')
for length in sen_counts.keys():
sen_rare = sen_rare_to_sen_count.get(length, 0)
sen_rare_pct = 100*sen_rare / total_sen if total_sen > 0 else 0
sen_unique = sen_unique_to_sen_count.get(length, 0)
sen_unique_pct = 100*sen_unique / total_sen if total_sen > 0 else 0
sen_risky = sen_rare + sen_unique
sen_risky_pct = 100*sen_risky / total_sen if total_sen > 0 else 0
f.write(
'\t'.join(
[str(length),
str(sen_rare),
str(sen_rare_pct),
str(sen_unique),
str(sen_unique_pct),
str(sen_risky),
str(sen_risky_pct)]) + '\n')
_, _, syn_length_to_combo_to_rare = util.mapShortestUniqueRareComboLengthToRecords(syn_records, len_to_syn_rare)
combos_tsv = path.join(output_dir, f'{prefix}_synthetic_rare_combos_by_length.tsv')
with open(combos_tsv, 'w') as f:
f.write('\t'.join(['combo_length', 'combo', 'record_id', 'syn_count', 'sen_count'])+'\n')
for length, combo_to_rare in syn_length_to_combo_to_rare.items():
for combo, rare_ids in combo_to_rare.items():
syn_count = len(rare_ids)
for rare_id in rare_ids:
sen_count = util.protect(sen_counts.get(length, {})[combo], reporting_resolution)
f.write(
'\t'.join(
[str(length),
util.comboToString(combo).replace(';', ' AND '),
str(rare_id),
str(syn_count),
str(sen_count)]) + '\n')
parameters_tsv = path.join(output_dir, f'{prefix}_parameters.tsv')
with open(parameters_tsv, 'w') as f:
f.write('\t'.join(['parameter', 'value'])+'\n')
f.write('\t'.join(['resolution', str(reporting_resolution)])+'\n')
f.write('\t'.join(['limit', str(reporting_length)])+'\n')
f.write('\t'.join(['total_sensitive_records', str(total_sen)])+'\n')
f.write('\t'.join(['unique_identifiable', str(unique_total)])+'\n')
f.write('\t'.join(['rare_identifiable', str(rare_total)])+'\n')
f.write('\t'.join(['risky_identifiable', str(risky_total)])+'\n')
f.write('\t'.join(['risky_identifiable_pct', str(risky_total_pct)])+'\n')
leakage_tsv = path.join(output_dir, f'{prefix}_synthetic_leakage_by_length.tsv')
leakage_svg = path.join(output_dir, f'{prefix}_synthetic_leakage_by_length.svg')
with open(leakage_tsv, 'w') as f:
f.write('\t'.join(['syn_combo_length', 'combo_count', 'leak_count', 'leak_proportion'])+'\n')
for length, leak_count in len_to_syn_leak.items():
combo_count = len_to_syn_count.get(length, 0)
leak_prop = leak_count/combo_count if combo_count > 0 else 0
f.write('\t'.join([str(length), str(combo_count), str(leak_count), str(leak_prop)])+'\n')
util.plotStats(
x_axis='syn_combo_length',
x_axis_title='Length of Synthetic Combination',
y_bar='combo_count',
y_bar_title='Count of Combinations',
y_line='leak_proportion',
y_line_title=f'Proportion of Leaked (<{reporting_resolution}) Combinations',
color='violet',
darker_color='darkviolet',
stats_tsv=leakage_tsv,
stats_svg=leakage_svg,
delimiter='\t',
style='whitegrid',
palette='magma')
compareDatasets(filtered_sen_counts, syn_counts, output_dir, prefix)
logging.info(
f'Evaluated {synthetic_microdata_path} vs {sensitive_microdata_path}, took {datetime.timedelta(seconds = time.time() - start_time)}s')
def compareDatasets(sensitive_length_to_combo_to_count, synthetic_length_to_combo_to_count, output_dir, prefix):
"""Evaluates the error in the synthetic microdata with respect to the sensitive microdata.
Produces output statistics and graphics binned by attribute count.
Args:
sensitive_length_to_combo_to_count: counts from sensitive microdata.
synthetic_length_to_combo_to_count: counts from synthetic microdata.
output_dir: where to save output statistics and graphics.
prefix: prefix to add to output files.
"""
all_count_length_preservation = []
max_syn_count = 0
all_combos = set()
for length, combo_to_count in sensitive_length_to_combo_to_count.items():
for combo in combo_to_count.keys():
all_combos.add((length, combo))
for length, combo_to_count in synthetic_length_to_combo_to_count.items():
for combo in combo_to_count.keys():
all_combos.add((length, combo))
tot = len(all_combos)
for i, (length, combo) in enumerate(all_combos):
if i % 10000 == 0:
logging.info(f'{100*i/tot:.1f}% through comparisons')
sen_count = sensitive_length_to_combo_to_count.get(length, {}).get(combo, 0)
syn_count = synthetic_length_to_combo_to_count.get(length, {}).get(combo, 0)
max_syn_count = max(syn_count, max_syn_count)
preservation = 0
preservation = syn_count / sen_count if sen_count > 0 else 0
if sen_count == 0:
logging.error(f'Error: For {combo}, synthetic count is {syn_count} but no sensitive count')
all_count_length_preservation.append((syn_count, length, preservation))
max_syn_count = max(syn_count, max_syn_count)
generateStatsAndGraphics(output_dir, max_syn_count, all_count_length_preservation, prefix)
def generateStatsAndGraphics(output_dir, max_syn_count, count_length_preservation, prefix):
"""Generates count error statistics for buckets of user-observed counts (post-filtering).
Outputs the files preservation_by_length and preservation_by_count tsv and svg files.
Args:
output_dir: the folder in which to output summary files.
max_syn_count: the maximum observed count of synthetic records matching a single attribute value.
count_length_preservation: list of (count, length, preservation) tuples for observed preservation of sensitive counts after filtering by a combo of length.
"""
sorted_counts = sorted(count_length_preservation, key=lambda x: x[0], reverse=False)
buckets = []
next_bucket = 10
while next_bucket < max_syn_count:
buckets.append(next_bucket)
next_bucket *= 2
buckets.append(next_bucket)
bucket_to_preservations = defaultdict(list)
bucket_to_counts = defaultdict(list)
bucket_to_lengths = defaultdict(list)
length_to_preservations = defaultdict(list)
length_to_counts = defaultdict(list)
for (count, length, preservation) in sorted_counts:
bucket = buckets[0]
for bi in range(len(buckets)-1, -1, -1):
if count > buckets[bi]:
bucket = buckets[bi+1]
break
bucket_to_preservations[bucket].append(preservation)
bucket_to_lengths[bucket].append(length)
bucket_to_counts[bucket].append(count)
length_to_counts[length].append(count)
length_to_preservations[length].append(preservation)
bucket_to_mean_count = {bucket: sum(counts)/len(counts) if len(counts) >
0 else 0 for bucket, counts in bucket_to_counts.items()}
bucket_to_mean_preservation = {bucket: sum(preservations) / len(preservations)
if len(preservations) > 0 else 0 for bucket,
preservations in bucket_to_preservations.items()}
bucket_to_mean_length = {bucket: sum(lengths)/len(lengths) if len(lengths) >
0 else 0 for bucket, lengths in bucket_to_lengths.items()}
counts_tsv = path.join(output_dir, f'{prefix}_preservation_by_count.tsv')
counts_svg = path.join(output_dir, f'{prefix}_preservation_by_count.svg')
with open(counts_tsv, 'w') as f:
f.write('\t'.join(['syn_count_bucket', 'mean_combo_count', 'mean_combo_length', 'count_preservation'])+'\n')
for bucket in reversed(buckets):
f.write(
'\t'.join(
[str(bucket),
str(bucket_to_mean_count.get(bucket, 0)),
str(bucket_to_mean_length.get(bucket, 0)),
str(bucket_to_mean_preservation.get(bucket, 0))]) + '\n')
util.plotStats(
x_axis='syn_count_bucket',
x_axis_title='Count of Filtered Synthetic Records',
y_bar='mean_combo_length',
y_bar_title='Mean Length of Combinations',
y_line='count_preservation',
y_line_title='Count Preservation (Synthetic/Sensitive)',
color='lightgreen',
darker_color='green',
stats_tsv=counts_tsv,
stats_svg=counts_svg,
delimiter='\t',
style='whitegrid',
palette='magma')
length_to_mean_preservation = {length: sum(preservations) / len(preservations)
if len(preservations) > 0 else 0 for length,
preservations in length_to_preservations.items()}
length_to_mean_count = {length: sum(counts)/len(counts) if len(counts) >
0 else 0 for length, counts in length_to_counts.items()}
lengths_tsv = path.join(output_dir, f'{prefix}_preservation_by_length.tsv')
lengths_svg = path.join(output_dir, f'{prefix}_preservation_by_length.svg')
with open(lengths_tsv, 'w') as f:
f.write('\t'.join(['syn_combo_length', 'mean_combo_count', 'count_preservation'])+'\n')
for length in sorted(length_to_preservations.keys()):
f.write(
'\t'.join(
[str(length),
str(length_to_mean_count.get(length, 0)),
str(length_to_mean_preservation.get(length, 0))]) + '\n')
util.plotStats(
x_axis='syn_combo_length',
x_axis_title='Length of Combination',
y_bar='mean_combo_count',
y_bar_title='Mean Synthetic Count',
y_line='count_preservation',
y_line_title='Count Preservation (Synthetic/Sensitive)',
color='cornflowerblue',
darker_color='mediumblue',
stats_tsv=lengths_tsv,
stats_svg=lengths_svg,
delimiter='\t',
style='whitegrid',
palette='magma')