openbooks-statements 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.
- openbooks_statements-0.1.0/.gitignore +49 -0
- openbooks_statements-0.1.0/LICENSE +202 -0
- openbooks_statements-0.1.0/PKG-INFO +45 -0
- openbooks_statements-0.1.0/README.md +27 -0
- openbooks_statements-0.1.0/hatch_build.py +19 -0
- openbooks_statements-0.1.0/models/roles_calibration.json +1 -0
- openbooks_statements-0.1.0/models/roles_head.pt +0 -0
- openbooks_statements-0.1.0/models/roles_model_card.json +13 -0
- openbooks_statements-0.1.0/pyproject.toml +48 -0
- openbooks_statements-0.1.0/src/openbooks_statements/__init__.py +4 -0
- openbooks_statements-0.1.0/src/openbooks_statements/_version.py +1 -0
- openbooks_statements-0.1.0/src/openbooks_statements/assets.py +28 -0
- openbooks_statements-0.1.0/src/openbooks_statements/calibration.py +47 -0
- openbooks_statements-0.1.0/src/openbooks_statements/encoder.py +151 -0
- openbooks_statements-0.1.0/src/openbooks_statements/frontend.py +17 -0
- openbooks_statements-0.1.0/src/openbooks_statements/gates.py +10 -0
- openbooks_statements-0.1.0/src/openbooks_statements/labels.py +3 -0
- openbooks_statements-0.1.0/src/openbooks_statements/parsers.py +185 -0
- openbooks_statements-0.1.0/src/openbooks_statements/pdf_backend.py +164 -0
- openbooks_statements-0.1.0/src/openbooks_statements/pdf_front.py +478 -0
- openbooks_statements-0.1.0/src/openbooks_statements/py.typed +0 -0
- openbooks_statements-0.1.0/src/openbooks_statements/roles.py +168 -0
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
# Python
|
|
2
|
+
__pycache__/
|
|
3
|
+
*.py[cod]
|
|
4
|
+
*.egg-info/
|
|
5
|
+
.pytest_cache/
|
|
6
|
+
.ruff_cache/
|
|
7
|
+
.mypy_cache/
|
|
8
|
+
.venv/
|
|
9
|
+
venv/
|
|
10
|
+
build/
|
|
11
|
+
dist/
|
|
12
|
+
|
|
13
|
+
# Secrets and local config
|
|
14
|
+
.env
|
|
15
|
+
.env.*
|
|
16
|
+
!.env.example
|
|
17
|
+
*.key
|
|
18
|
+
*.pem
|
|
19
|
+
.claude/
|
|
20
|
+
|
|
21
|
+
# Downloaded third-party datasets (licences: some are NC / share-alike; never redistribute)
|
|
22
|
+
data/raw/
|
|
23
|
+
data/download_log.json
|
|
24
|
+
data/*_stdout.txt
|
|
25
|
+
|
|
26
|
+
# Generated training data (rebuilt by scripts/train_roles.py)
|
|
27
|
+
data/*.pkl
|
|
28
|
+
|
|
29
|
+
# Internal-only weights: pretrained on CORD (CC BY-SA), must not be shipped
|
|
30
|
+
models/jepa_encoder.pt
|
|
31
|
+
# Experimental runs (leave-one-family-out etc.); only the promoted model in models/ is tracked
|
|
32
|
+
models/fam_*/
|
|
33
|
+
models/*_bak*/
|
|
34
|
+
|
|
35
|
+
# Web app (web/)
|
|
36
|
+
node_modules/
|
|
37
|
+
.next/
|
|
38
|
+
out/
|
|
39
|
+
*.tsbuildinfo
|
|
40
|
+
next-env.d.ts
|
|
41
|
+
|
|
42
|
+
# OS / editor
|
|
43
|
+
.DS_Store
|
|
44
|
+
Thumbs.db
|
|
45
|
+
.vscode/
|
|
46
|
+
.idea/
|
|
47
|
+
*.log
|
|
48
|
+
|
|
49
|
+
.hypothesis/
|
|
@@ -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.
|
|
@@ -0,0 +1,45 @@
|
|
|
1
|
+
Metadata-Version: 2.5
|
|
2
|
+
Name: openbooks-statements
|
|
3
|
+
Version: 0.1.0
|
|
4
|
+
Summary: Open Books providers for bank statements: bank CSV, OFX, MT940, PDF and scans (OCR), optional typed-head cross-check
|
|
5
|
+
License-Expression: Apache-2.0
|
|
6
|
+
License-File: LICENSE
|
|
7
|
+
Requires-Python: >=3.11
|
|
8
|
+
Requires-Dist: openbooks-core[xlsx]<0.2,>=0.1.0
|
|
9
|
+
Provides-Extra: pdf
|
|
10
|
+
Requires-Dist: pypdfium2>=4.30; extra == 'pdf'
|
|
11
|
+
Provides-Extra: scan
|
|
12
|
+
Requires-Dist: easyocr; extra == 'scan'
|
|
13
|
+
Requires-Dist: numpy; extra == 'scan'
|
|
14
|
+
Requires-Dist: pillow; extra == 'scan'
|
|
15
|
+
Requires-Dist: pypdfium2>=4.30; extra == 'scan'
|
|
16
|
+
Requires-Dist: torch>=2.6; extra == 'scan'
|
|
17
|
+
Description-Content-Type: text/markdown
|
|
18
|
+
|
|
19
|
+
# openbooks-statements
|
|
20
|
+
|
|
21
|
+
Providers for bank statements, registered with `openbooks-core`.
|
|
22
|
+
|
|
23
|
+
| Input | Install | How it is proven |
|
|
24
|
+
|---|---|---|
|
|
25
|
+
| bank CSV (`bank_csv.v1`), OFX, MT940 | `pip install openbooks-statements` (pure Python) | opening + movements = closing, running balance ties on every row |
|
|
26
|
+
| PDF with a text layer | `pip install "openbooks-statements[pdf]"` (adds pypdfium2) | same chain, with page and bounding box evidence per field |
|
|
27
|
+
| scans and photos | `pip install "openbooks-statements[scan]"` (adds EasyOCR, PyTorch, Pillow) | same chain; OCR dates are also checked to be in order, and low OCR confidence refuses |
|
|
28
|
+
|
|
29
|
+
Without the `[pdf]` extra a PDF is not a refusal about the document: the engine reports that pypdfium2 is missing (exit code 3).
|
|
30
|
+
PDFs are read with pypdfium2 (Apache-2.0 / BSD-3, wrapping Google's PDFium) and images with Pillow; nothing in this package is copyleft.
|
|
31
|
+
A page whose text layer is rotated (`/Rotate`) is refused rather than guessed; scans are rendered and read by OCR.
|
|
32
|
+
|
|
33
|
+
## Typed heads (`--model`)
|
|
34
|
+
|
|
35
|
+
`python -m openbooks_core run statement.pdf --model` cross-checks the column assignment with a small calibrated model. The heads
|
|
36
|
+
can only add refusals (`LOW_CONFIDENCE`); the balance gate always has the last word. The weights (`roles_head.pt`, about 19 MB)
|
|
37
|
+
and their calibration ship inside the wheel under `openbooks_statements/data/`, and the calibration names the sha256 of the exact
|
|
38
|
+
weights it was fitted on: a mismatch refuses to load. A released document records both hashes in `file.model`.
|
|
39
|
+
`OPENBOOKS_MODELS_DIR` overrides the location. Weights are loaded with `torch.load(..., weights_only=True)` (PyTorch 2.6 or newer).
|
|
40
|
+
|
|
41
|
+
All figures for the heads are on synthetic data. No accuracy claim has been measured on real client documents.
|
|
42
|
+
|
|
43
|
+
## Not here
|
|
44
|
+
|
|
45
|
+
Training, the synthetic data generator, calibration fitting and payout-to-bank-credit matching are not part of this package.
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
# openbooks-statements
|
|
2
|
+
|
|
3
|
+
Providers for bank statements, registered with `openbooks-core`.
|
|
4
|
+
|
|
5
|
+
| Input | Install | How it is proven |
|
|
6
|
+
|---|---|---|
|
|
7
|
+
| bank CSV (`bank_csv.v1`), OFX, MT940 | `pip install openbooks-statements` (pure Python) | opening + movements = closing, running balance ties on every row |
|
|
8
|
+
| PDF with a text layer | `pip install "openbooks-statements[pdf]"` (adds pypdfium2) | same chain, with page and bounding box evidence per field |
|
|
9
|
+
| scans and photos | `pip install "openbooks-statements[scan]"` (adds EasyOCR, PyTorch, Pillow) | same chain; OCR dates are also checked to be in order, and low OCR confidence refuses |
|
|
10
|
+
|
|
11
|
+
Without the `[pdf]` extra a PDF is not a refusal about the document: the engine reports that pypdfium2 is missing (exit code 3).
|
|
12
|
+
PDFs are read with pypdfium2 (Apache-2.0 / BSD-3, wrapping Google's PDFium) and images with Pillow; nothing in this package is copyleft.
|
|
13
|
+
A page whose text layer is rotated (`/Rotate`) is refused rather than guessed; scans are rendered and read by OCR.
|
|
14
|
+
|
|
15
|
+
## Typed heads (`--model`)
|
|
16
|
+
|
|
17
|
+
`python -m openbooks_core run statement.pdf --model` cross-checks the column assignment with a small calibrated model. The heads
|
|
18
|
+
can only add refusals (`LOW_CONFIDENCE`); the balance gate always has the last word. The weights (`roles_head.pt`, about 19 MB)
|
|
19
|
+
and their calibration ship inside the wheel under `openbooks_statements/data/`, and the calibration names the sha256 of the exact
|
|
20
|
+
weights it was fitted on: a mismatch refuses to load. A released document records both hashes in `file.model`.
|
|
21
|
+
`OPENBOOKS_MODELS_DIR` overrides the location. Weights are loaded with `torch.load(..., weights_only=True)` (PyTorch 2.6 or newer).
|
|
22
|
+
|
|
23
|
+
All figures for the heads are on synthetic data. No accuracy claim has been measured on real client documents.
|
|
24
|
+
|
|
25
|
+
## Not here
|
|
26
|
+
|
|
27
|
+
Training, the synthetic data generator, calibration fitting and payout-to-bank-credit matching are not part of this package.
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
"""Ships the model assets in both targets. The source of truth is models/ at the repo root; the sdist carries a copy under
|
|
2
|
+
models/ so a wheel built from the sdist gets the same weights, calibration and model card (the hash binding needs the pair)."""
|
|
3
|
+
from pathlib import Path
|
|
4
|
+
|
|
5
|
+
from hatchling.builders.hooks.plugin.interface import BuildHookInterface
|
|
6
|
+
|
|
7
|
+
ASSETS = ("roles_head.pt", "roles_calibration.json", "roles_model_card.json")
|
|
8
|
+
|
|
9
|
+
|
|
10
|
+
class ModelAssets(BuildHookInterface):
|
|
11
|
+
def initialize(self, version, build_data):
|
|
12
|
+
root = Path(self.root)
|
|
13
|
+
src = root / "models" if (root / "models").is_dir() else root / ".." / ".." / ".." / "models"
|
|
14
|
+
for name in ASSETS:
|
|
15
|
+
path = src / name
|
|
16
|
+
if not path.is_file():
|
|
17
|
+
raise FileNotFoundError(f"model asset missing: {path}")
|
|
18
|
+
dest = f"models/{name}" if self.target_name == "sdist" else f"openbooks_statements/data/{name}"
|
|
19
|
+
build_data["force_include"][str(path.resolve())] = dest
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"role": {"temperature": 2.019380728929039, "qhat": 0.9976497888565063, "alpha": 0.001}, "vtype": {"temperature": 1.0769541104862101, "qhat": 0.9351925849914551, "alpha": 0.001}, "weights_sha256": "d101e39610b74fa3195399a9e4aed89dad2916c2ac874284b08a38b6a0d31056"}
|
|
Binary file
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
{
|
|
2
|
+
"model": "RoleNet (typed heads: column role + value type) over a LayoutLM-style encoder",
|
|
3
|
+
"parameters": 18878602,
|
|
4
|
+
"weights_mb_int8_at_rest": 19.04,
|
|
5
|
+
"weights_sha256": "d101e39610b74fa3195399a9e4aed89dad2916c2ac874284b08a38b6a0d31056",
|
|
6
|
+
"initialisation": "random; no pretrained weights",
|
|
7
|
+
"training_data": "synthetic statements only; no third-party dataset and no real client document",
|
|
8
|
+
"calibration": {
|
|
9
|
+
"method": "temperature scaling + split conformal",
|
|
10
|
+
"alpha": 0.001
|
|
11
|
+
},
|
|
12
|
+
"role": "cross-check only: the heads can add refusals, never cause a release"
|
|
13
|
+
}
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
[build-system]
|
|
2
|
+
requires = ["hatchling>=1.27"]
|
|
3
|
+
build-backend = "hatchling.build"
|
|
4
|
+
|
|
5
|
+
[project]
|
|
6
|
+
name = "openbooks-statements"
|
|
7
|
+
dynamic = ["version"]
|
|
8
|
+
description = "Open Books providers for bank statements: bank CSV, OFX, MT940, PDF and scans (OCR), optional typed-head cross-check"
|
|
9
|
+
readme = "README.md"
|
|
10
|
+
requires-python = ">=3.11"
|
|
11
|
+
license = "Apache-2.0"
|
|
12
|
+
license-files = ["LICENSE"]
|
|
13
|
+
# The base install is pure Python: bank CSV, OFX and MT940. PDF and scans are extras so the PDF library and PyTorch
|
|
14
|
+
# dependencies are opt-in.
|
|
15
|
+
dependencies = ["openbooks-core[xlsx]>=0.1.0,<0.2"]
|
|
16
|
+
|
|
17
|
+
[project.optional-dependencies]
|
|
18
|
+
pdf = ["pypdfium2>=4.30"]
|
|
19
|
+
scan = ["openbooks-statements[pdf]", "easyocr", "torch>=2.6", "pillow", "numpy"]
|
|
20
|
+
|
|
21
|
+
[project.entry-points."openbooks.layouts"]
|
|
22
|
+
"bank_csv.v1" = "openbooks_statements.parsers:parse_bank_csv"
|
|
23
|
+
|
|
24
|
+
[project.entry-points."openbooks.sniffers"]
|
|
25
|
+
"ofx.v1" = "openbooks_statements.parsers:parse_ofx"
|
|
26
|
+
"mt940.v1" = "openbooks_statements.parsers:parse_mt940"
|
|
27
|
+
|
|
28
|
+
[project.entry-points."openbooks.frontends"]
|
|
29
|
+
document = "openbooks_statements.frontend:FRONTEND"
|
|
30
|
+
|
|
31
|
+
[project.entry-points."openbooks.gates"]
|
|
32
|
+
"statement:ocr_dates" = "openbooks_statements.gates:ocr_dates"
|
|
33
|
+
|
|
34
|
+
[tool.hatch.version]
|
|
35
|
+
path = "src/openbooks_statements/_version.py"
|
|
36
|
+
|
|
37
|
+
[tool.hatch.build.targets.wheel]
|
|
38
|
+
packages = ["src/openbooks_statements"]
|
|
39
|
+
|
|
40
|
+
# The weights and their calibration travel together in one wheel so the hash binding can never see a mismatched pair.
|
|
41
|
+
# Source of truth stays models/ at the repo root. Only the weights, their calibration and the model card ship (hatch_build.py),
|
|
42
|
+
# in the wheel and in the sdist, so a wheel built from the sdist is the same.
|
|
43
|
+
[tool.hatch.build.targets.wheel.hooks.custom]
|
|
44
|
+
|
|
45
|
+
[tool.hatch.build.targets.sdist.hooks.custom]
|
|
46
|
+
|
|
47
|
+
[tool.hatch.build.targets.sdist]
|
|
48
|
+
include = ["LICENSE", "src/openbooks_statements", "README.md", "pyproject.toml", "hatch_build.py"]
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
__version__ = "0.1.0"
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
"""Where the model assets are, with no torch import.
|
|
2
|
+
|
|
3
|
+
Order: the directory named by OPENBOOKS_MODELS_DIR, then the assets packaged inside this distribution (data/), then the
|
|
4
|
+
repo's models/ directory when running from a source checkout. The calibration is bound to the weights by sha256 and the
|
|
5
|
+
loader refuses a mismatch, so a stray or older file elsewhere can never be used silently."""
|
|
6
|
+
from __future__ import annotations
|
|
7
|
+
|
|
8
|
+
import os
|
|
9
|
+
from importlib import resources
|
|
10
|
+
from pathlib import Path
|
|
11
|
+
|
|
12
|
+
WEIGHTS_NAME, CALIBRATION_NAME = "roles_head.pt", "roles_calibration.json"
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
def models_dir() -> Path:
|
|
16
|
+
env = os.environ.get("OPENBOOKS_MODELS_DIR")
|
|
17
|
+
if env:
|
|
18
|
+
return Path(env)
|
|
19
|
+
packaged = Path(str(resources.files("openbooks_statements").joinpath("data")))
|
|
20
|
+
if (packaged / WEIGHTS_NAME).exists():
|
|
21
|
+
return packaged
|
|
22
|
+
for parent in Path(__file__).resolve().parents:
|
|
23
|
+
if (parent / "models" / WEIGHTS_NAME).exists():
|
|
24
|
+
return parent / "models"
|
|
25
|
+
return packaged # absent: load() reports that there are no weights
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
MODELS_DIR = models_dir()
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
"""Calibration and abstention for the typed heads at inference: temperature-scaled probabilities and split-conformal sets.
|
|
2
|
+
|
|
3
|
+
A token is only trusted when its conformal set holds exactly one class. Anything else abstains, which the pipeline reports
|
|
4
|
+
as LOW_CONFIDENCE. Coverage (the true class is in the set) is guaranteed at 1 - alpha only if calibration and test data are
|
|
5
|
+
exchangeable; on layouts unlike the calibration set it can fail. The numbers (temperature, qhat, alpha) ship in
|
|
6
|
+
roles_calibration.json, bound to the exact weights by hash. Fitting them is not part of this package.
|
|
7
|
+
"""
|
|
8
|
+
from __future__ import annotations
|
|
9
|
+
|
|
10
|
+
import json
|
|
11
|
+
import math
|
|
12
|
+
from dataclasses import dataclass
|
|
13
|
+
from pathlib import Path
|
|
14
|
+
|
|
15
|
+
import numpy as np
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
def softmax(z: np.ndarray) -> np.ndarray:
|
|
19
|
+
z = z - z.max(axis=-1, keepdims=True)
|
|
20
|
+
e = np.exp(z)
|
|
21
|
+
return e / e.sum(axis=-1, keepdims=True)
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
def prediction_set(probs: np.ndarray, qhat: float) -> list[int]:
|
|
25
|
+
return [i for i, p in enumerate(probs) if 1.0 - p <= qhat]
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
@dataclass(frozen=True)
|
|
29
|
+
class Calibrator:
|
|
30
|
+
temperature: float = 1.0
|
|
31
|
+
qhat: float = float("inf")
|
|
32
|
+
alpha: float = 0.05
|
|
33
|
+
|
|
34
|
+
def probs(self, logits: np.ndarray) -> np.ndarray:
|
|
35
|
+
return softmax(np.asarray(logits, dtype=np.float64) / self.temperature)
|
|
36
|
+
|
|
37
|
+
def sets(self, logits: np.ndarray) -> list[list[int]]:
|
|
38
|
+
return [prediction_set(p, self.qhat) for p in self.probs(logits)]
|
|
39
|
+
|
|
40
|
+
def save(self, path: Path) -> None:
|
|
41
|
+
path.write_text(json.dumps({"temperature": self.temperature, "qhat": None if math.isinf(self.qhat) else self.qhat,
|
|
42
|
+
"alpha": self.alpha}), encoding="utf-8")
|
|
43
|
+
|
|
44
|
+
@classmethod
|
|
45
|
+
def load(cls, path: Path) -> "Calibrator":
|
|
46
|
+
d = json.loads(path.read_text(encoding="utf-8"))
|
|
47
|
+
return cls(d["temperature"], float("inf") if d["qhat"] is None else d["qhat"], d["alpha"])
|
|
@@ -0,0 +1,151 @@
|
|
|
1
|
+
"""Layout encoder shared by the typed heads: hashed word and shape embeddings plus 2D box embeddings.
|
|
2
|
+
|
|
3
|
+
Release-one size: about 19M parameters (budget 25M; int8 at rest is about 1 MB per million). Trained from scratch on
|
|
4
|
+
synthetic statements only. Nothing here loads a pretrained checkpoint.
|
|
5
|
+
"""
|
|
6
|
+
from __future__ import annotations
|
|
7
|
+
|
|
8
|
+
import re
|
|
9
|
+
import zlib
|
|
10
|
+
|
|
11
|
+
import torch
|
|
12
|
+
from torch import nn
|
|
13
|
+
|
|
14
|
+
NB, NS, MAXLEN, D, LAYERS, HEADS = 16384, 512, 512, 384, 6, 6
|
|
15
|
+
MONEY = re.compile(r"^\(?-?\d[\d,]*\.\d{2}\)?(?:Dr|Cr|DR|CR)?$")
|
|
16
|
+
NOT_MONEY, MAX_RANK, MAX_N = 5, 4, 6
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
def column_heads(boxes: list[list[int]], hashes: list[int], money: list[bool]) -> list[list[int]]:
|
|
20
|
+
"""For each amount token, the hashed words (up to 3) of the nearest non-amount line above it in the same x-range.
|
|
21
|
+
|
|
22
|
+
On a statement that is the column header ('Withdrawal Amt.', 'Credit'); the model learns what those words mean. Nothing
|
|
23
|
+
here knows any header vocabulary, so a new header wording is just a new hash, not a rule to maintain.
|
|
24
|
+
"""
|
|
25
|
+
import numpy as np
|
|
26
|
+
|
|
27
|
+
heads = [[0, 0, 0] for _ in boxes]
|
|
28
|
+
if not any(money):
|
|
29
|
+
return heads
|
|
30
|
+
b, mon = np.asarray(boxes), np.asarray(money)
|
|
31
|
+
for i in np.nonzero(mon)[0]:
|
|
32
|
+
ok = ~mon & (np.minimum(b[:, 2], b[i, 2]) - np.maximum(b[:, 0], b[i, 0]) > 0) & (b[:, 3] <= b[i, 1])
|
|
33
|
+
idx = np.nonzero(ok)[0]
|
|
34
|
+
if not len(idx):
|
|
35
|
+
continue
|
|
36
|
+
near = idx[np.argmax(b[idx, 3])]
|
|
37
|
+
h = max(4, b[near, 3] - b[near, 1])
|
|
38
|
+
line = idx[np.abs((b[idx, 1] + b[idx, 3]) / 2 - (b[near, 1] + b[near, 3]) / 2) <= 0.6 * h]
|
|
39
|
+
ids = [hashes[k] for k in line[np.argsort(b[line, 0])][:3]]
|
|
40
|
+
heads[i] = ids + [0] * (3 - len(ids))
|
|
41
|
+
return heads
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
def row_features(boxes: list[list[int]], money: list[bool]) -> list[list[int]]:
|
|
45
|
+
"""Per token (rank from the right among the amount tokens of its row, number of amounts in the row).
|
|
46
|
+
|
|
47
|
+
A statement row reads description ... [debit] [credit] balance, so the balance is always rank 0. This is geometry plus
|
|
48
|
+
the surface money test only: no header words and no rule-based column assignment, so it stays independent of the extractor.
|
|
49
|
+
"""
|
|
50
|
+
feats = [[NOT_MONEY, 0] for _ in boxes]
|
|
51
|
+
order = sorted((i for i, m in enumerate(money) if m), key=lambda i: boxes[i][1] + boxes[i][3])
|
|
52
|
+
rows: list[list[int]] = []
|
|
53
|
+
for i in order:
|
|
54
|
+
cy, h = (boxes[i][1] + boxes[i][3]) / 2, max(4, boxes[i][3] - boxes[i][1])
|
|
55
|
+
if rows and abs(cy - sum((boxes[k][1] + boxes[k][3]) / 2 for k in rows[-1]) / len(rows[-1])) <= 0.6 * h:
|
|
56
|
+
rows[-1].append(i)
|
|
57
|
+
else:
|
|
58
|
+
rows.append([i])
|
|
59
|
+
for row in rows:
|
|
60
|
+
for rank, i in enumerate(sorted(row, key=lambda k: -boxes[k][2])):
|
|
61
|
+
feats[i] = [min(rank, MAX_RANK), min(len(row), MAX_N)]
|
|
62
|
+
return feats
|
|
63
|
+
|
|
64
|
+
|
|
65
|
+
def h_word(w: str) -> int:
|
|
66
|
+
return zlib.crc32(w.lower().encode()) % (NB - 2) + 2
|
|
67
|
+
|
|
68
|
+
|
|
69
|
+
def h_shape(w: str) -> int:
|
|
70
|
+
s = "".join("d" if c.isdigit() else "A" if c.isupper() else "a" if c.isalpha() else c for c in w)
|
|
71
|
+
return zlib.crc32((re.sub(r"(.)\1+", r"\1", s) + str(min(len(w), 12))).encode()) % (NS - 1) + 1
|
|
72
|
+
|
|
73
|
+
|
|
74
|
+
class Enc(nn.Module):
|
|
75
|
+
"""Token + shape + 2D box embeddings (LayoutLM-style), LAYERS x width D transformer."""
|
|
76
|
+
|
|
77
|
+
def __init__(self, d: int = D, layers: int = LAYERS, heads: int = HEADS):
|
|
78
|
+
super().__init__()
|
|
79
|
+
self.word, self.shape = nn.Embedding(NB, d, padding_idx=0), nn.Embedding(NS, d, padding_idx=0)
|
|
80
|
+
self.xe, self.ye, self.we, self.he = (nn.Embedding(1001, d) for _ in range(4))
|
|
81
|
+
self.pos = nn.Embedding(MAXLEN, d)
|
|
82
|
+
self.rank, self.nmon = nn.Embedding(NOT_MONEY + 1, d), nn.Embedding(MAX_N + 1, d)
|
|
83
|
+
layer = nn.TransformerEncoderLayer(d, heads, 4 * d, 0.1, activation="gelu", batch_first=True, norm_first=True)
|
|
84
|
+
self.tf = nn.TransformerEncoder(layer, layers, enable_nested_tensor=False)
|
|
85
|
+
self.norm = nn.LayerNorm(d)
|
|
86
|
+
|
|
87
|
+
def forward(self, W, S, X, pad, R, H):
|
|
88
|
+
pos = torch.arange(W.shape[1], device=W.device)
|
|
89
|
+
wd, ht = (X[..., 2] - X[..., 0]).clamp(0, 1000), (X[..., 3] - X[..., 1]).clamp(0, 1000)
|
|
90
|
+
x = (self.word(W) + self.shape(S) + self.xe(X[..., 0]) + self.xe(X[..., 2]) + self.ye(X[..., 1])
|
|
91
|
+
+ self.ye(X[..., 3]) + self.we(wd) + self.he(ht) + self.pos(pos) + self.rank(R[..., 0]) + self.nmon(R[..., 1])
|
|
92
|
+
+ self.word(H).sum(2))
|
|
93
|
+
return self.norm(self.tf(x, src_key_padding_mask=pad))
|
|
94
|
+
|
|
95
|
+
|
|
96
|
+
class RoleNet(nn.Module):
|
|
97
|
+
"""Encoder plus two typed heads: the column role of a token and the value type it holds."""
|
|
98
|
+
|
|
99
|
+
def __init__(self, n_roles: int, n_types: int):
|
|
100
|
+
super().__init__()
|
|
101
|
+
self.enc = Enc()
|
|
102
|
+
self.role, self.vtype = nn.Linear(D, n_roles), nn.Linear(D, n_types)
|
|
103
|
+
|
|
104
|
+
def forward(self, W, S, X, pad, R, H):
|
|
105
|
+
h = self.enc(W, S, X, pad, R, H)
|
|
106
|
+
return self.role(h), self.vtype(h)
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
def windows(words) -> list[list[int]]:
|
|
110
|
+
"""Index windows of at most MAXLEN tokens per page, in reading order (page, y, x)."""
|
|
111
|
+
out: list[list[int]] = []
|
|
112
|
+
by_page: dict[int, list[int]] = {}
|
|
113
|
+
for i, w in enumerate(words):
|
|
114
|
+
by_page.setdefault(w.page, []).append(i)
|
|
115
|
+
for idx in by_page.values():
|
|
116
|
+
idx.sort(key=lambda i: (round(words[i].cy / 4), words[i].x0))
|
|
117
|
+
out += [idx[k:k + MAXLEN] for k in range(0, len(idx), MAXLEN)]
|
|
118
|
+
return out
|
|
119
|
+
|
|
120
|
+
|
|
121
|
+
def tensors(words, idx: list[int], dev):
|
|
122
|
+
"""Collate one window into (W, S, X, pad, R) batch-of-one tensors, boxes scaled to 0..1000 by page size."""
|
|
123
|
+
ws = [words[i] for i in idx]
|
|
124
|
+
clamp = lambda v: min(1000, max(0, int(v))) # noqa: E731
|
|
125
|
+
boxes = [[clamp(w.x0 * 1000 / w.pw), clamp(w.y0 * 1000 / w.ph), clamp(w.x1 * 1000 / w.pw), clamp(w.y1 * 1000 / w.ph)] for w in ws]
|
|
126
|
+
W = torch.tensor([[h_word(w.text) for w in ws]], device=dev)
|
|
127
|
+
S = torch.tensor([[h_shape(w.text) for w in ws]], device=dev)
|
|
128
|
+
money = [bool(MONEY.match(w.text)) for w in ws]
|
|
129
|
+
R = torch.tensor([row_features(boxes, money)], device=dev)
|
|
130
|
+
H = torch.tensor([column_heads(boxes, [h_word(w.text) for w in ws], money)], device=dev)
|
|
131
|
+
return W, S, torch.tensor([boxes], device=dev), torch.zeros(1, len(ws), dtype=torch.bool, device=dev), R, H
|
|
132
|
+
|
|
133
|
+
|
|
134
|
+
def n_params(net: nn.Module) -> int:
|
|
135
|
+
return sum(p.numel() for p in net.parameters())
|
|
136
|
+
|
|
137
|
+
|
|
138
|
+
def quantize_state(state: dict, min_numel: int = 4096) -> dict:
|
|
139
|
+
"""int8 at rest: per-row symmetric scales for every large 2D weight (embeddings and linears), fp16 for the rest."""
|
|
140
|
+
out = {}
|
|
141
|
+
for k, v in state.items():
|
|
142
|
+
if v.ndim == 2 and v.numel() >= min_numel:
|
|
143
|
+
scale = (v.abs().amax(dim=1, keepdim=True) / 127).clamp(min=1e-8)
|
|
144
|
+
out[k] = {"q": torch.round(v / scale).clamp(-127, 127).to(torch.int8), "s": scale.squeeze(1).half()}
|
|
145
|
+
else:
|
|
146
|
+
out[k] = v.half()
|
|
147
|
+
return out
|
|
148
|
+
|
|
149
|
+
|
|
150
|
+
def dequantize_state(blob: dict) -> dict:
|
|
151
|
+
return {k: (v["q"].float() * v["s"].float().unsqueeze(1) if isinstance(v, dict) else v.float()) for k, v in blob.items()}
|