whowhen 0.1.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- whowhen-0.1.0/LICENSE +202 -0
- whowhen-0.1.0/NOTICE +2 -0
- whowhen-0.1.0/PKG-INFO +262 -0
- whowhen-0.1.0/README.md +225 -0
- whowhen-0.1.0/pyproject.toml +52 -0
- whowhen-0.1.0/setup.cfg +4 -0
- whowhen-0.1.0/src/whowhen/__init__.py +16 -0
- whowhen-0.1.0/src/whowhen/_stats.py +49 -0
- whowhen-0.1.0/src/whowhen/audit.py +268 -0
- whowhen-0.1.0/src/whowhen/cli.py +115 -0
- whowhen-0.1.0/src/whowhen/clock.py +127 -0
- whowhen-0.1.0/src/whowhen/columns.py +111 -0
- whowhen-0.1.0/src/whowhen/demo.py +99 -0
- whowhen-0.1.0/src/whowhen/overlap.py +280 -0
- whowhen-0.1.0/src/whowhen/pairs.py +201 -0
- whowhen-0.1.0/src/whowhen/panel.py +123 -0
- whowhen-0.1.0/src/whowhen/report.py +231 -0
- whowhen-0.1.0/src/whowhen/scorers.py +144 -0
- whowhen-0.1.0/src/whowhen.egg-info/PKG-INFO +262 -0
- whowhen-0.1.0/src/whowhen.egg-info/SOURCES.txt +26 -0
- whowhen-0.1.0/src/whowhen.egg-info/dependency_links.txt +1 -0
- whowhen-0.1.0/src/whowhen.egg-info/entry_points.txt +2 -0
- whowhen-0.1.0/src/whowhen.egg-info/requires.txt +19 -0
- whowhen-0.1.0/src/whowhen.egg-info/top_level.txt +1 -0
- whowhen-0.1.0/tests/test_audit.py +198 -0
- whowhen-0.1.0/tests/test_columns.py +59 -0
- whowhen-0.1.0/tests/test_overlap.py +106 -0
- whowhen-0.1.0/tests/test_report.py +180 -0
whowhen-0.1.0/LICENSE
ADDED
|
@@ -0,0 +1,202 @@
|
|
|
1
|
+
|
|
2
|
+
Apache License
|
|
3
|
+
Version 2.0, January 2004
|
|
4
|
+
http://www.apache.org/licenses/
|
|
5
|
+
|
|
6
|
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
|
7
|
+
|
|
8
|
+
1. Definitions.
|
|
9
|
+
|
|
10
|
+
"License" shall mean the terms and conditions for use, reproduction,
|
|
11
|
+
and distribution as defined by Sections 1 through 9 of this document.
|
|
12
|
+
|
|
13
|
+
"Licensor" shall mean the copyright owner or entity authorized by
|
|
14
|
+
the copyright owner that is granting the License.
|
|
15
|
+
|
|
16
|
+
"Legal Entity" shall mean the union of the acting entity and all
|
|
17
|
+
other entities that control, are controlled by, or are under common
|
|
18
|
+
control with that entity. For the purposes of this definition,
|
|
19
|
+
"control" means (i) the power, direct or indirect, to cause the
|
|
20
|
+
direction or management of such entity, whether by contract or
|
|
21
|
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
|
22
|
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
|
23
|
+
|
|
24
|
+
"You" (or "Your") shall mean an individual or Legal Entity
|
|
25
|
+
exercising permissions granted by this License.
|
|
26
|
+
|
|
27
|
+
"Source" form shall mean the preferred form for making modifications,
|
|
28
|
+
including but not limited to software source code, documentation
|
|
29
|
+
source, and configuration files.
|
|
30
|
+
|
|
31
|
+
"Object" form shall mean any form resulting from mechanical
|
|
32
|
+
transformation or translation of a Source form, including but
|
|
33
|
+
not limited to compiled object code, generated documentation,
|
|
34
|
+
and conversions to other media types.
|
|
35
|
+
|
|
36
|
+
"Work" shall mean the work of authorship, whether in Source or
|
|
37
|
+
Object form, made available under the License, as indicated by a
|
|
38
|
+
copyright notice that is included in or attached to the work
|
|
39
|
+
(an example is provided in the Appendix below).
|
|
40
|
+
|
|
41
|
+
"Derivative Works" shall mean any work, whether in Source or Object
|
|
42
|
+
form, that is based on (or derived from) the Work and for which the
|
|
43
|
+
editorial revisions, annotations, elaborations, or other modifications
|
|
44
|
+
represent, as a whole, an original work of authorship. For the purposes
|
|
45
|
+
of this License, Derivative Works shall not include works that remain
|
|
46
|
+
separable from, or merely link (or bind by name) to the interfaces of,
|
|
47
|
+
the Work and Derivative Works thereof.
|
|
48
|
+
|
|
49
|
+
"Contribution" shall mean any work of authorship, including
|
|
50
|
+
the original version of the Work and any modifications or additions
|
|
51
|
+
to that Work or Derivative Works thereof, that is intentionally
|
|
52
|
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
|
53
|
+
or by an individual or Legal Entity authorized to submit on behalf of
|
|
54
|
+
the copyright owner. For the purposes of this definition, "submitted"
|
|
55
|
+
means any form of electronic, verbal, or written communication sent
|
|
56
|
+
to the Licensor or its representatives, including but not limited to
|
|
57
|
+
communication on electronic mailing lists, source code control systems,
|
|
58
|
+
and issue tracking systems that are managed by, or on behalf of, the
|
|
59
|
+
Licensor for the purpose of discussing and improving the Work, but
|
|
60
|
+
excluding communication that is conspicuously marked or otherwise
|
|
61
|
+
designated in writing by the copyright owner as "Not a Contribution."
|
|
62
|
+
|
|
63
|
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
|
64
|
+
on behalf of whom a Contribution has been received by Licensor and
|
|
65
|
+
subsequently incorporated within the Work.
|
|
66
|
+
|
|
67
|
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
|
68
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
69
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
70
|
+
copyright license to reproduce, prepare Derivative Works of,
|
|
71
|
+
publicly display, publicly perform, sublicense, and distribute the
|
|
72
|
+
Work and such Derivative Works in Source or Object form.
|
|
73
|
+
|
|
74
|
+
3. Grant of Patent License. Subject to the terms and conditions of
|
|
75
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
76
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
77
|
+
(except as stated in this section) patent license to make, have made,
|
|
78
|
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
|
79
|
+
where such license applies only to those patent claims licensable
|
|
80
|
+
by such Contributor that are necessarily infringed by their
|
|
81
|
+
Contribution(s) alone or by combination of their Contribution(s)
|
|
82
|
+
with the Work to which such Contribution(s) was submitted. If You
|
|
83
|
+
institute patent litigation against any entity (including a
|
|
84
|
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
|
85
|
+
or a Contribution incorporated within the Work constitutes direct
|
|
86
|
+
or contributory patent infringement, then any patent licenses
|
|
87
|
+
granted to You under this License for that Work shall terminate
|
|
88
|
+
as of the date such litigation is filed.
|
|
89
|
+
|
|
90
|
+
4. Redistribution. You may reproduce and distribute copies of the
|
|
91
|
+
Work or Derivative Works thereof in any medium, with or without
|
|
92
|
+
modifications, and in Source or Object form, provided that You
|
|
93
|
+
meet the following conditions:
|
|
94
|
+
|
|
95
|
+
(a) You must give any other recipients of the Work or
|
|
96
|
+
Derivative Works a copy of this License; and
|
|
97
|
+
|
|
98
|
+
(b) You must cause any modified files to carry prominent notices
|
|
99
|
+
stating that You changed the files; and
|
|
100
|
+
|
|
101
|
+
(c) You must retain, in the Source form of any Derivative Works
|
|
102
|
+
that You distribute, all copyright, patent, trademark, and
|
|
103
|
+
attribution notices from the Source form of the Work,
|
|
104
|
+
excluding those notices that do not pertain to any part of
|
|
105
|
+
the Derivative Works; and
|
|
106
|
+
|
|
107
|
+
(d) If the Work includes a "NOTICE" text file as part of its
|
|
108
|
+
distribution, then any Derivative Works that You distribute must
|
|
109
|
+
include a readable copy of the attribution notices contained
|
|
110
|
+
within such NOTICE file, excluding those notices that do not
|
|
111
|
+
pertain to any part of the Derivative Works, in at least one
|
|
112
|
+
of the following places: within a NOTICE text file distributed
|
|
113
|
+
as part of the Derivative Works; within the Source form or
|
|
114
|
+
documentation, if provided along with the Derivative Works; or,
|
|
115
|
+
within a display generated by the Derivative Works, if and
|
|
116
|
+
wherever such third-party notices normally appear. The contents
|
|
117
|
+
of the NOTICE file are for informational purposes only and
|
|
118
|
+
do not modify the License. You may add Your own attribution
|
|
119
|
+
notices within Derivative Works that You distribute, alongside
|
|
120
|
+
or as an addendum to the NOTICE text from the Work, provided
|
|
121
|
+
that such additional attribution notices cannot be construed
|
|
122
|
+
as modifying the License.
|
|
123
|
+
|
|
124
|
+
You may add Your own copyright statement to Your modifications and
|
|
125
|
+
may provide additional or different license terms and conditions
|
|
126
|
+
for use, reproduction, or distribution of Your modifications, or
|
|
127
|
+
for any such Derivative Works as a whole, provided Your use,
|
|
128
|
+
reproduction, and distribution of the Work otherwise complies with
|
|
129
|
+
the conditions stated in this License.
|
|
130
|
+
|
|
131
|
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
|
132
|
+
any Contribution intentionally submitted for inclusion in the Work
|
|
133
|
+
by You to the Licensor shall be under the terms and conditions of
|
|
134
|
+
this License, without any additional terms or conditions.
|
|
135
|
+
Notwithstanding the above, nothing herein shall supersede or modify
|
|
136
|
+
the terms of any separate license agreement you may have executed
|
|
137
|
+
with Licensor regarding such Contributions.
|
|
138
|
+
|
|
139
|
+
6. Trademarks. This License does not grant permission to use the trade
|
|
140
|
+
names, trademarks, service marks, or product names of the Licensor,
|
|
141
|
+
except as required for reasonable and customary use in describing the
|
|
142
|
+
origin of the Work and reproducing the content of the NOTICE file.
|
|
143
|
+
|
|
144
|
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
|
145
|
+
agreed to in writing, Licensor provides the Work (and each
|
|
146
|
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
|
147
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
148
|
+
implied, including, without limitation, any warranties or conditions
|
|
149
|
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
|
150
|
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
|
151
|
+
appropriateness of using or redistributing the Work and assume any
|
|
152
|
+
risks associated with Your exercise of permissions under this License.
|
|
153
|
+
|
|
154
|
+
8. Limitation of Liability. In no event and under no legal theory,
|
|
155
|
+
whether in tort (including negligence), contract, or otherwise,
|
|
156
|
+
unless required by applicable law (such as deliberate and grossly
|
|
157
|
+
negligent acts) or agreed to in writing, shall any Contributor be
|
|
158
|
+
liable to You for damages, including any direct, indirect, special,
|
|
159
|
+
incidental, or consequential damages of any character arising as a
|
|
160
|
+
result of this License or out of the use or inability to use the
|
|
161
|
+
Work (including but not limited to damages for loss of goodwill,
|
|
162
|
+
work stoppage, computer failure or malfunction, or any and all
|
|
163
|
+
other commercial damages or losses), even if such Contributor
|
|
164
|
+
has been advised of the possibility of such damages.
|
|
165
|
+
|
|
166
|
+
9. Accepting Warranty or Additional Liability. While redistributing
|
|
167
|
+
the Work or Derivative Works thereof, You may choose to offer,
|
|
168
|
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
|
169
|
+
or other liability obligations and/or rights consistent with this
|
|
170
|
+
License. However, in accepting such obligations, You may act only
|
|
171
|
+
on Your own behalf and on Your sole responsibility, not on behalf
|
|
172
|
+
of any other Contributor, and only if You agree to indemnify,
|
|
173
|
+
defend, and hold each Contributor harmless for any liability
|
|
174
|
+
incurred by, or claims asserted against, such Contributor by reason
|
|
175
|
+
of your accepting any such warranty or additional liability.
|
|
176
|
+
|
|
177
|
+
END OF TERMS AND CONDITIONS
|
|
178
|
+
|
|
179
|
+
APPENDIX: How to apply the Apache License to your work.
|
|
180
|
+
|
|
181
|
+
To apply the Apache License to your work, attach the following
|
|
182
|
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
|
183
|
+
replaced with your own identifying information. (Don't include
|
|
184
|
+
the brackets!) The text should be enclosed in the appropriate
|
|
185
|
+
comment syntax for the file format. We also recommend that a
|
|
186
|
+
file or class name and description of purpose be included on the
|
|
187
|
+
same "printed page" as the copyright notice for easier
|
|
188
|
+
identification within third-party archives.
|
|
189
|
+
|
|
190
|
+
Copyright [yyyy] [name of copyright owner]
|
|
191
|
+
|
|
192
|
+
Licensed under the Apache License, Version 2.0 (the "License");
|
|
193
|
+
you may not use this file except in compliance with the License.
|
|
194
|
+
You may obtain a copy of the License at
|
|
195
|
+
|
|
196
|
+
http://www.apache.org/licenses/LICENSE-2.0
|
|
197
|
+
|
|
198
|
+
Unless required by applicable law or agreed to in writing, software
|
|
199
|
+
distributed under the License is distributed on an "AS IS" BASIS,
|
|
200
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
201
|
+
See the License for the specific language governing permissions and
|
|
202
|
+
limitations under the License.
|
whowhen-0.1.0/NOTICE
ADDED
whowhen-0.1.0/PKG-INFO
ADDED
|
@@ -0,0 +1,262 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: whowhen
|
|
3
|
+
Version: 0.1.0
|
|
4
|
+
Summary: Audit panel data for identity and timing signals a model can exploit without predicting anything
|
|
5
|
+
Author: Andrew Hendel
|
|
6
|
+
License: Apache-2.0
|
|
7
|
+
Project-URL: Repository, https://github.com/mojolearn/whowhen
|
|
8
|
+
Keywords: data leakage,evaluation,panel data,dataset audit,time series
|
|
9
|
+
Classifier: Development Status :: 3 - Alpha
|
|
10
|
+
Classifier: Intended Audience :: Science/Research
|
|
11
|
+
Classifier: License :: OSI Approved :: Apache Software License
|
|
12
|
+
Classifier: Programming Language :: Python :: 3
|
|
13
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
14
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
15
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
16
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
17
|
+
Classifier: Programming Language :: Python :: 3.14
|
|
18
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
19
|
+
Requires-Python: >=3.10
|
|
20
|
+
Description-Content-Type: text/markdown
|
|
21
|
+
License-File: LICENSE
|
|
22
|
+
License-File: NOTICE
|
|
23
|
+
Requires-Dist: numpy>=1.24
|
|
24
|
+
Requires-Dist: pandas>=2.0
|
|
25
|
+
Requires-Dist: scikit-learn>=1.3
|
|
26
|
+
Requires-Dist: joblib>=1.3
|
|
27
|
+
Requires-Dist: threadpoolctl>=3.1
|
|
28
|
+
Requires-Dist: mojolearn>=0.8.66; (sys_platform == "darwin" and platform_machine == "arm64") or (sys_platform == "linux" and platform_machine == "x86_64")
|
|
29
|
+
Requires-Dist: catboost>=1.2; (sys_platform != "darwin" or platform_machine != "arm64") and (sys_platform != "linux" or platform_machine != "x86_64")
|
|
30
|
+
Provides-Extra: parquet
|
|
31
|
+
Requires-Dist: pyarrow>=14.0; extra == "parquet"
|
|
32
|
+
Provides-Extra: test
|
|
33
|
+
Requires-Dist: pytest>=7; extra == "test"
|
|
34
|
+
Requires-Dist: catboost>=1.2; extra == "test"
|
|
35
|
+
Requires-Dist: pyarrow>=14.0; extra == "test"
|
|
36
|
+
Dynamic: license-file
|
|
37
|
+
|
|
38
|
+
# whowhen
|
|
39
|
+
|
|
40
|
+
Check a table of repeated observations per unit (patients, machines,
|
|
41
|
+
loans, users, vessels) for two signals a model can exploit without
|
|
42
|
+
predicting anything:
|
|
43
|
+
|
|
44
|
+
- **who**: can a row be matched to the unit it came from?
|
|
45
|
+
- **when**: can two rows of one unit be put in time order?
|
|
46
|
+
|
|
47
|
+
If the answer to either is yes, a model evaluated on your data may be
|
|
48
|
+
scoring well by recognizing units or reading the clock.
|
|
49
|
+
|
|
50
|
+
## Install
|
|
51
|
+
|
|
52
|
+
```
|
|
53
|
+
pip install whowhen
|
|
54
|
+
```
|
|
55
|
+
|
|
56
|
+
## Use
|
|
57
|
+
|
|
58
|
+
```python
|
|
59
|
+
import pandas as pd
|
|
60
|
+
import whowhen as ww
|
|
61
|
+
|
|
62
|
+
df = pd.read_parquet("stays.parquet")
|
|
63
|
+
print(ww.report(df, unit="patient_id", time="hour", event="sepsis"))
|
|
64
|
+
```
|
|
65
|
+
|
|
66
|
+
That is the whole interface.
|
|
67
|
+
|
|
68
|
+
| Setting | Required | Default | What it is |
|
|
69
|
+
|---|---|---|---|
|
|
70
|
+
| `df` | yes | | the table, one row per observation |
|
|
71
|
+
| `unit` | yes | | the column naming the unit |
|
|
72
|
+
| `time` | yes | | the column giving when the row was observed |
|
|
73
|
+
| `event` | no | none | a column that is nonzero where the event happens |
|
|
74
|
+
| `features` | no | every numeric column | the columns to check: a list, or a dict of named lists to compare side by side |
|
|
75
|
+
| `horizon` | no | none | for forecasting tasks: how far ahead the event is predicted, in the time column's scale (`12`, or `"12h"` for datetimes) |
|
|
76
|
+
| `repeats` | no | 4 | how many times every check is run |
|
|
77
|
+
| `random` | no | `False` | draw held-out units at random instead of the fixed groups |
|
|
78
|
+
| `n_pairs` | no | 20000 | the most row pairs each run uses; smaller tables use what they have |
|
|
79
|
+
|
|
80
|
+
What runs depends on what you give:
|
|
81
|
+
|
|
82
|
+
| You give | You get |
|
|
83
|
+
|---|---|
|
|
84
|
+
| `unit` and `time` | the identity and timing audits, the timing placebo, the individual-or-type test, the audits by short and long gap, the per-column ranking |
|
|
85
|
+
| plus `event` | the overlap check, the chronological control, the time-cut split, the clock baseline |
|
|
86
|
+
| plus `horizon` | those three run as a forecast: "does the event come within this long after the row?" |
|
|
87
|
+
|
|
88
|
+
A check that cannot run is listed at the end of the report with the
|
|
89
|
+
reason.
|
|
90
|
+
|
|
91
|
+
To compare choices made before modeling, such as raw values against a
|
|
92
|
+
carried-forward or imputed version, put both versions in the table and
|
|
93
|
+
name them:
|
|
94
|
+
|
|
95
|
+
```python
|
|
96
|
+
ww.report(df, unit="patient_id", time="hour",
|
|
97
|
+
features={"raw": raw_cols, "carried_forward": filled_cols})
|
|
98
|
+
```
|
|
99
|
+
|
|
100
|
+
To see the library work before pointing it at your own data, run it on
|
|
101
|
+
a synthetic table where the right answers are known:
|
|
102
|
+
|
|
103
|
+
```
|
|
104
|
+
whowhen demo
|
|
105
|
+
```
|
|
106
|
+
|
|
107
|
+
## What the report contains
|
|
108
|
+
|
|
109
|
+
**Identity audit and timing audit.** Both train a classifier on pairs
|
|
110
|
+
of rows from units with no recorded event, built so that the two
|
|
111
|
+
classes look the same unless the signal is there. The gap between the
|
|
112
|
+
two rows of a pair is chosen from the data and stated in the report.
|
|
113
|
+
|
|
114
|
+
| Audit | The pair | Label | AUC 0.5 means |
|
|
115
|
+
|---|---|---|---|
|
|
116
|
+
| identity | two rows a set time apart | same unit, or two different units seen at the same two times | rows carry nothing that says which unit produced them |
|
|
117
|
+
| timing | two rows of one unit a set time apart | which row is the later one | a unit's rows carry nothing that says when they were recorded |
|
|
118
|
+
|
|
119
|
+
Units are split into training and held-out sides before any pair is
|
|
120
|
+
drawn, so the AUC is measured on units the classifier never saw. The
|
|
121
|
+
interval comes from resampling held-out units, pooled over repeats.
|
|
122
|
+
|
|
123
|
+
**Timing placebo.** The timing audit again, with the later row taken
|
|
124
|
+
from a different unit observed at that time. If this score is near 0.5,
|
|
125
|
+
the timing signal is each unit's own history. If it is near the timing
|
|
126
|
+
audit's score, the signal is change shared by all units, such as
|
|
127
|
+
calendar drift or a change in how every unit is recorded.
|
|
128
|
+
|
|
129
|
+
**Individual or type.** The identity audit is reported twice. With
|
|
130
|
+
`random` impostors the different unit is any other unit, so a high
|
|
131
|
+
score can mean "these two rows are from different kinds of unit" (a
|
|
132
|
+
tug and a tanker). With `nearest` impostors the different unit is one
|
|
133
|
+
of the units most similar to the first, by per-unit column averages.
|
|
134
|
+
If the score stays high, individual units are recognizable. If it
|
|
135
|
+
falls toward 0.5, the columns identify the kind of unit, not the
|
|
136
|
+
individual.
|
|
137
|
+
|
|
138
|
+
**Short and long gaps.** Both audits again, once with the two rows of
|
|
139
|
+
a pair in the shorter half of the gap range and once in the longer
|
|
140
|
+
half. If identity stays high at the long gap, units stay recognizable
|
|
141
|
+
as time passes.
|
|
142
|
+
|
|
143
|
+
**Columns.** For each column: the audit using that column alone, and
|
|
144
|
+
how much the full score drops when that column is removed.
|
|
145
|
+
|
|
146
|
+
**Overlap check.** Needs `event`. It scores the same test rows twice
|
|
147
|
+
with the same number of training rows per class: once with the test
|
|
148
|
+
units held out of training, once with other rows of those units
|
|
149
|
+
swapped in. The gap between the two scores is how much sharing units
|
|
150
|
+
between training and test is worth on your task.
|
|
151
|
+
|
|
152
|
+
**Chronological control.** The same check, testing the later half of
|
|
153
|
+
each test unit's rows and letting training see the earlier half: does
|
|
154
|
+
a unit's own past help predict its own future?
|
|
155
|
+
|
|
156
|
+
**Time-cut split.** The same check with a single cut in the time
|
|
157
|
+
column: test rows come after it, training rows before it. When the
|
|
158
|
+
time column is a shared calendar, this is the usual split by date, and
|
|
159
|
+
the gap shows how much unit overlap such a split leaves in place.
|
|
160
|
+
|
|
161
|
+
**Clock baseline.** Many event tasks can be scored well above chance
|
|
162
|
+
from one number: how long the unit has been observed. This fits the
|
|
163
|
+
same classifier on the clock alone, on your columns alone, and on
|
|
164
|
+
both, with units held out, and reports what the columns add once the
|
|
165
|
+
clock is known.
|
|
166
|
+
|
|
167
|
+
## What the numbers do not mean
|
|
168
|
+
|
|
169
|
+
An audit AUC above 0.5 says the signal is present in the columns. It
|
|
170
|
+
does not say your model uses it; the overlap check tests that.
|
|
171
|
+
|
|
172
|
+
**An AUC near 0.5 does not certify an evaluation.** It says this
|
|
173
|
+
classifier, on these columns, found nothing. Leakage through the
|
|
174
|
+
label, through preprocessing fitted on test rows, or through columns
|
|
175
|
+
not in the table is outside what these checks see.
|
|
176
|
+
|
|
177
|
+
## When it applies
|
|
178
|
+
|
|
179
|
+
- many units (hundreds or more), each observed more than once;
|
|
180
|
+
- for the identity audit, several units observed at the same time
|
|
181
|
+
values, since the different-unit pair is matched on time;
|
|
182
|
+
- with `horizon`, an event column flagged on the row where the event
|
|
183
|
+
happens, not on every row of the unit.
|
|
184
|
+
|
|
185
|
+
It does not apply to tables with one row per example and no repeated
|
|
186
|
+
unit.
|
|
187
|
+
|
|
188
|
+
## Repeats and repeatability
|
|
189
|
+
|
|
190
|
+
`repeats=N` is one number: every check runs N times. Units fall into
|
|
191
|
+
N + 1 fixed groups, by a hash of the unit identifier, and each run
|
|
192
|
+
holds out a different group:
|
|
193
|
+
|
|
194
|
+
| `repeats` | Each run tests |
|
|
195
|
+
|---|---|
|
|
196
|
+
| 1 | one half of the units |
|
|
197
|
+
| 2 | one third |
|
|
198
|
+
| 3 | one quarter |
|
|
199
|
+
| 4 (default) | one fifth |
|
|
200
|
+
|
|
201
|
+
More repeats give a steadier answer, but each run then tests a smaller
|
|
202
|
+
share of the units, so very high values suit only large tables.
|
|
203
|
+
|
|
204
|
+
By default the same call on the same table gives the same numbers
|
|
205
|
+
every time. `random=True` draws the held-out units and the pairs at
|
|
206
|
+
random instead, differently on each run, and the report states the
|
|
207
|
+
seed that run used.
|
|
208
|
+
|
|
209
|
+
## The classifier
|
|
210
|
+
|
|
211
|
+
The checks use [mojolearn](https://github.com/mojolearn/mojolearn)'s
|
|
212
|
+
`GradientBoosting` with oblivious (symmetric) trees. mojolearn ships
|
|
213
|
+
wheels for macOS on Apple silicon and Linux on x86-64. On other
|
|
214
|
+
platforms, including Windows, CatBoost is installed and used instead,
|
|
215
|
+
with the same settings. The report names the classifier that ran.
|
|
216
|
+
|
|
217
|
+
Independent runs are spread across all cores. Set the `WHOWHEN_JOBS`
|
|
218
|
+
environment variable to use fewer.
|
|
219
|
+
|
|
220
|
+
## Command line
|
|
221
|
+
|
|
222
|
+
```
|
|
223
|
+
whowhen report rows.parquet --unit patient_id --time hour --event sepsis
|
|
224
|
+
whowhen demo
|
|
225
|
+
```
|
|
226
|
+
|
|
227
|
+
To use a report in a script or a scheduled job, say what counts as a
|
|
228
|
+
failure:
|
|
229
|
+
|
|
230
|
+
```
|
|
231
|
+
whowhen report rows.parquet --unit patient_id --time hour --event sepsis \
|
|
232
|
+
--fail-if "overlap_gap > 0.02" --fail-if "clock_gain < 0.01"
|
|
233
|
+
```
|
|
234
|
+
|
|
235
|
+
The command exits with 1 if any condition holds, and with 2 if a
|
|
236
|
+
condition names a check that did not run. You choose the conditions;
|
|
237
|
+
there are no built-in thresholds.
|
|
238
|
+
|
|
239
|
+
| Name | The number |
|
|
240
|
+
|---|---|
|
|
241
|
+
| `identity`, `identity_nearest` | identity AUC against any other unit, and against the most similar units |
|
|
242
|
+
| `timing`, `placebo` | timing AUC, and its placebo |
|
|
243
|
+
| `overlap_gap`, `chronological_gap`, `time_cut_gap` | overlap AUC minus disjoint AUC under each split |
|
|
244
|
+
| `clock` | AUC of the clock alone |
|
|
245
|
+
| `clock_gain` | what the columns add once the clock is known |
|
|
246
|
+
|
|
247
|
+
With several named feature sets, an audit's number is the highest
|
|
248
|
+
across them. A high identity AUC is normal for many tables and is not
|
|
249
|
+
by itself a failure; the gaps say whether it matters for your task.
|
|
250
|
+
|
|
251
|
+
## Citing
|
|
252
|
+
|
|
253
|
+
The method is described in:
|
|
254
|
+
|
|
255
|
+
> Andrew Hendel. Auditing Identity and Timing Signals in Operational
|
|
256
|
+
> Logs. 2026. Under review.
|
|
257
|
+
|
|
258
|
+
See `CITATION.cff`.
|
|
259
|
+
|
|
260
|
+
## License
|
|
261
|
+
|
|
262
|
+
Apache License 2.0. See `LICENSE`.
|