-
Notifications
You must be signed in to change notification settings - Fork 1
/
bootstrap_trials.py
195 lines (172 loc) · 7.01 KB
/
bootstrap_trials.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
import os
import pickle
import datetime
from dateutil.relativedelta import relativedelta
import yaml
import pandas as pd
import sqlalchemy
import RecallAdjuster as ra
NUM_TRIALS = 5
bootstrap_weights = [
{'W': 0.66, 'B': 0.27, 'H': 0.07}
# {'W': 0.62, 'B': 0.27, 'H': 0.11}, # current
# {'W': 0.58, 'B': 0.27, 'H': 0.15},
# {'W': 0.54, 'B': 0.27, 'H': 0.19},
# {'W': 0.50, 'B': 0.27, 'H': 0.23},
# {'W': 0.46, 'B': 0.27, 'H': 0.27},
# {'W': 0.42, 'B': 0.27, 'H': 0.31},
# {'W': 0.38, 'B': 0.27, 'H': 0.35},
# {'W': 0.34, 'B': 0.27, 'H': 0.39}
]
append_to = 'boostrap_hisp_frac_test.pkl'
export_file = 'boostrap_hisp_frac_test_new.pkl'
base = datetime.datetime.strptime('2018-04-01', '%Y-%m-%d')
date_pairs = []
for x in range(9,-1,-1):
date_pairs.append(
(
(base - relativedelta(months=4*x) - relativedelta(years=1)).strftime('%Y-%m-%d'),
(base - relativedelta(months=4*x) - relativedelta(years=1)).strftime('%Y-%m-%d')
)
)
date_pairs.append(
(
(base - relativedelta(months=4*x) - relativedelta(years=1)).strftime('%Y-%m-%d'),
(base - relativedelta(months=4*x)).strftime('%Y-%m-%d')
)
)
def connect(poolclass=sqlalchemy.pool.QueuePool):
with open(os.path.join(os.path.join('../..', 'config'), 'db_default_profile.yaml')) as fd:
config = yaml.load(fd)
dburl = sqlalchemy.engine.url.URL(
"postgres",
host=config["host"],
username=config["user"],
database=config["db"],
password=config["pass"],
port=config["port"],
)
return sqlalchemy.create_engine(dburl, poolclass=poolclass)
conn = connect()
all_ts = []
for bs_wts in bootstrap_weights:
for i in range(NUM_TRIALS):
print('starting trial %s of %s with weights %s...' % (i, NUM_TRIALS, bs_wts))
myRA = ra.RecallAdjuster(
engine=conn,
pg_role='kit',
schema='kit_bias_class_test',
experiment_hashes='09b3bcab5a6e1eb1c712571f6a5abb75',
date_pairs=date_pairs,
list_sizes=[500, 1000],
#entity_demos='joco',
entity_demos='kit_bias_class_test.entity_demos',
demo_col='race_3way',
bootstrap_weights=bs_wts
)
sql = """
WITH mg_rns AS (
SELECT *,
row_number() OVER (PARTITION BY train_end_time, list_size, metric, parameter ORDER BY base_value DESC, base_max_recall_ratio ASC, RANDOM()) AS rn_base,
row_number() OVER (PARTITION BY train_end_time, list_size, metric, parameter ORDER BY adj_value DESC, adj_max_recall_ratio ASC, RANDOM()) AS rn_adj
FROM kit_bias_class_test.model_adjustment_results_race_3way
WHERE past_train_end_time = train_end_time
)
, base_mgs AS (
SELECT * FROM mg_rns WHERE rn_base = 1
)
, adj_mgs AS (
SELECT * FROM mg_rns WHERE rn_adj = 1
)
-- Simple model selection on last time period, baseline with no recall adjustments
SELECT 'Best Unadjusted Metric - Unadjusted'::VARCHAR(128) AS strategy,
r.train_end_time, r.past_train_end_time,
r.list_size, r.metric, r.parameter,
r.base_value AS value,
r.base_max_recall_ratio AS max_recall_ratio,
r.base_recall_w_to_b AS recall_w_to_b,
r.base_recall_w_to_h AS recall_w_to_h,
r.base_recall_b_to_h AS recall_b_to_h,
r.base_frac_w AS frac_w,
r.base_frac_b AS frac_b,
r.base_frac_h AS frac_h
FROM kit_bias_class_test.model_adjustment_results_race_3way r
JOIN base_mgs b
ON r.model_group_id = b.model_group_id
AND r.past_train_end_time = b.train_end_time
AND r.list_size = b.list_size
AND r.metric = b.metric
AND r.parameter = b.parameter
WHERE r.train_end_time > r.past_train_end_time
UNION ALL
-- Model selection on last time before adjustment, with adjustment applied
SELECT 'Best Unadjusted Metric - Adjusted'::VARCHAR(128) AS strategy,
r.train_end_time, r.past_train_end_time,
r.list_size, r.metric, r.parameter,
r.adj_value AS value,
r.adj_max_recall_ratio AS max_recall_ratio,
r.adj_recall_w_to_b AS recall_w_to_b,
r.adj_recall_w_to_h AS recall_w_to_h,
r.adj_recall_b_to_h AS recall_b_to_h,
r.adj_frac_w AS frac_w,
r.adj_frac_b AS frac_b,
r.adj_frac_h AS frac_h
FROM kit_bias_class_test.model_adjustment_results_race_3way r
JOIN base_mgs b
ON r.model_group_id = b.model_group_id
AND r.past_train_end_time = b.train_end_time
AND r.list_size = b.list_size
AND r.metric = b.metric
AND r.parameter = b.parameter
WHERE r.train_end_time > r.past_train_end_time
UNION ALL
-- Model selection on last time after adjustment, with adjustment applied
SELECT 'Best Adjusted Metric - Adjusted'::VARCHAR(128) AS strategy,
r.train_end_time, r.past_train_end_time,
r.list_size, r.metric, r.parameter,
r.adj_value AS value,
r.adj_max_recall_ratio AS max_recall_ratio,
r.adj_recall_w_to_b AS recall_w_to_b,
r.adj_recall_w_to_h AS recall_w_to_h,
r.adj_recall_b_to_h AS recall_b_to_h,
r.adj_frac_w AS frac_w,
r.adj_frac_b AS frac_b,
r.adj_frac_h AS frac_h
FROM kit_bias_class_test.model_adjustment_results_race_3way r
JOIN adj_mgs b
ON r.model_group_id = b.model_group_id
AND r.past_train_end_time = b.train_end_time
AND r.list_size = b.list_size
AND r.metric = b.metric
AND r.parameter = b.parameter
WHERE r.train_end_time > r.past_train_end_time
UNION ALL
-- Composite model (no decoupled models)
SELECT 'Composite Model - Adjusted'::VARCHAR(128) AS strategy,
train_end_time, past_train_end_time,
list_size, metric, parameter,
value,
max_recall_ratio,
recall_w_to_b,
recall_w_to_h,
recall_b_to_h,
frac_w,
frac_b,
frac_h
FROM kit_bias_class_test.composite_results_race_3way
WHERE train_end_time > past_train_end_time
;
"""
new_ts = pd.read_sql(sql, conn)
new_ts['bootstrap_frac_w'] = bs_wts['W']
new_ts['bootstrap_frac_b'] = bs_wts['B']
new_ts['bootstrap_frac_h'] = bs_wts['H']
new_ts['bootstrap_trial'] = i
all_ts.append(new_ts)
if append_to:
with open(append_to, 'rb') as f:
old_res = pickle.load(f)
all_ts.append(old_res)
results = pd.concat(all_ts)
with open(export_file, 'wb') as f:
pickle.dump(results, f)