├── .coveragerc ├── .flake8 ├── .gitignore ├── .isort.cfg ├── .pre-commit-config.yaml ├── CITATION.cff ├── CONTRIBUTING.md ├── LICENSE ├── Makefile ├── Pipfile ├── Pipfile.lock ├── README.md ├── docs └── assets │ ├── banner.png │ └── datasets.b9bdb1e1.png ├── examples └── 01_hapi_intro.ipynb ├── hapi ├── __init__.py ├── convert.py ├── dataset.py └── version.py ├── pyproject.toml └── setup.py /.coveragerc: -------------------------------------------------------------------------------- 1 | [run] 2 | branch = True 3 | source = hapi 4 | 5 | [report] 6 | exclude_lines = 7 | if self.debug: 8 | pragma: no cover 9 | raise NotImplementedError 10 | raise NotImplemented 11 | if __name__ == .__main__.: 12 | ignore_errors = True 13 | omit = 14 | tests/* 15 | setup.py -------------------------------------------------------------------------------- /.flake8: -------------------------------------------------------------------------------- 1 | # This is our code-style check. We currently allow the following exceptions: 2 | # - E731: do not assign a lambda expression, use a def 3 | # - W503: line break before binary operator 4 | # - E741: do not use variables named 'l', 'O', or 'I' 5 | # - E203: whitespace before ':' 6 | [flake8] 7 | count = True 8 | max-line-length = 88 9 | statistics = True 10 | ignore = E731,W503,E741,E203 11 | exclude = setup.py -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | dcbench-config.yaml 6 | data 7 | 8 | # C extensions 9 | *.so 10 | *.DS_Store 11 | 12 | # Distribution / packaging 13 | .Python 14 | build/ 15 | develop-eggs/ 16 | dist/ 17 | downloads/ 18 | eggs/ 19 | .eggs/ 20 | lib/ 21 | lib64/ 22 | parts/ 23 | sdist/ 24 | var/ 25 | tmp/ 26 | wheels/ 27 | pip-wheel-metadata/ 28 | share/python-wheels/ 29 | *.egg-info/ 30 | .installed.cfg 31 | *.egg 32 | MANIFEST 33 | 34 | # PyInstaller 35 | # Usually these files are written by a python script from a template 36 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 37 | *.manifest 38 | *.spec 39 | 40 | # Installer logs 41 | pip-log.txt 42 | pip-delete-this-directory.txt 43 | 44 | # Unit test / coverage reports 45 | htmlcov/ 46 | .tox/ 47 | .nox/ 48 | .coverage 49 | .coverage.* 50 | .cache 51 | nosetests.xml 52 | coverage.xml 53 | *.cover 54 | *.py,cover 55 | .hypothesis/ 56 | .pytest_cache/ 57 | 58 | # Translations 59 | *.mo 60 | *.pot 61 | 62 | # Django stuff: 63 | *.log 64 | local_settings.py 65 | db.sqlite3 66 | db.sqlite3-journal 67 | 68 | # Flask stuff: 69 | instance/ 70 | .webassets-cache 71 | 72 | # Scrapy stuff: 73 | .scrapy 74 | 75 | # Sphinx documentation 76 | docs/_build/ 77 | 78 | # PyBuilder 79 | target/ 80 | 81 | # Jupyter Notebook 82 | .ipynb_checkpoints 83 | 84 | # IPython 85 | profile_default/ 86 | ipython_config.py 87 | 88 | # pyenv 89 | .python-version 90 | 91 | # pipenv 92 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 93 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 94 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 95 | # install all needed dependencies. 96 | #Pipfile.lock 97 | 98 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow 99 | __pypackages__/ 100 | 101 | # Celery stuff 102 | celerybeat-schedule 103 | celerybeat.pid 104 | 105 | # SageMath parsed files 106 | *.sage.py 107 | 108 | # Environments 109 | .env 110 | .venv 111 | env/ 112 | venv/ 113 | ENV/ 114 | env.bak/ 115 | venv.bak/ 116 | 117 | # Spyder project settings 118 | .spyderproject 119 | .spyproject 120 | 121 | # Rope project settings 122 | .ropeproject 123 | 124 | # mkdocs documentation 125 | /site 126 | 127 | # mypy 128 | .mypy_cache/ 129 | .dmypy.json 130 | dmypy.json 131 | 132 | # Pyre type checker 133 | .pyre/ 134 | 135 | # Other stuff 136 | .vscode/ 137 | tmp/ -------------------------------------------------------------------------------- /.isort.cfg: -------------------------------------------------------------------------------- 1 | [settings] 2 | multi_line_output = 3 3 | include_trailing_comma = True 4 | force_grid_wrap = 0 5 | use_parentheses = True 6 | ensure_newline_before_comments = True 7 | line_length = 88 -------------------------------------------------------------------------------- /.pre-commit-config.yaml: -------------------------------------------------------------------------------- 1 | repos: 2 | - repo: https://github.com/timothycrosley/isort 3 | rev: 5.7.0 4 | hooks: 5 | - id: isort 6 | - repo: https://github.com/psf/black 7 | rev: 20.8b1 8 | hooks: 9 | - id: black 10 | language_version: python3 11 | - repo: https://gitlab.com/pycqa/flake8 12 | rev: 3.8.4 13 | hooks: 14 | - id: flake8 -------------------------------------------------------------------------------- /CITATION.cff: -------------------------------------------------------------------------------- 1 | cff-version: 1.2.0 2 | message: "If you use this benchmark, please cite it as below." 3 | authors: 4 | - family-names: Chen 5 | given-names: Lingjiao 6 | - family-names: Eyuboglu 7 | given-names: Sabri 8 | orcid: "https://orcid.org/0000-0002-8412-0266" 9 | - family-names: Jin 10 | given-names: Zhihua 11 | - family-names: Ré 12 | given-names: Christopher 13 | - family-names: Zaharia 14 | given-names: Matei 15 | - family-names: Zou 16 | given-names: James 17 | title: "hapi" 18 | version: 1.0.0 19 | doi: 10.5281/zenodo.1234 20 | date-released: 2021-11-29 21 | url: "https://github.com/lchen001/HAPI" 22 | -------------------------------------------------------------------------------- /CONTRIBUTING.md: -------------------------------------------------------------------------------- 1 | # Contributing to hapi 2 | 3 | We welcome contributions of all kinds: code, documentation, feedback and support. If 4 | you use hapi in your work (blogs posts, research, company) and find it 5 | useful, spread the word! 6 | 7 | This contribution borrows from and is heavily inspired by [Huggingface transformers](https://github.com/huggingface/transformers). 8 | 9 | ## How to contribute 10 | 11 | There are 4 ways you can contribute: 12 | * Issues: raising bugs, suggesting new features 13 | * Fixes: resolving outstanding bugs 14 | * Features: contributing new features 15 | * Documentation: contributing documentation or examples 16 | 17 | ## Submitting a new issue or feature request 18 | 19 | Do your best to follow these guidelines when submitting an issue or a feature 20 | request. It will make it easier for us to give feedback and move your request forward. 21 | 22 | ### Bugs 23 | 24 | First, we would really appreciate it if you could **make sure the bug was not 25 | already reported** (use the search bar on Github under Issues). 26 | 27 | If you didn't find anything, please use the bug issue template to file a Github issue. 28 | 29 | 30 | ### Features 31 | 32 | A world-class feature request addresses the following points: 33 | 34 | 1. Motivation first: 35 | * Is it related to a problem/frustration with the library? If so, please explain 36 | why. Providing a code snippet that demonstrates the problem is best. 37 | * Is it related to something you would need for a project? We'd love to hear 38 | about it! 39 | * Is it something you worked on and think could benefit the community? 40 | Awesome! Tell us what problem it solved for you. 41 | 2. Write a *full paragraph* describing the feature; 42 | 3. Provide a **code snippet** that demonstrates its future use; 43 | 4. In case this is related to a paper, please attach a link; 44 | 5. Attach any additional information (drawings, screenshots, etc.) you think may help. 45 | 46 | If your issue is well written we're already 80% of the way there by the time you 47 | post it. 48 | 49 | ## Contributing (Pull Requests) 50 | 51 | Before writing code, we strongly advise you to search through the existing PRs or 52 | issues to make sure that nobody is already working on the same thing. If you are 53 | unsure, it is always a good idea to open an issue to get some feedback. 54 | 55 | You will need basic `git` proficiency to be able to contribute to 56 | `hapi`. `git` is not the easiest tool to use but it has the greatest 57 | manual. Type `git --help` in a shell and enjoy. If you prefer books, [Pro 58 | Git](https://git-scm.com/book/en/v2) is a very good reference. 59 | 60 | Follow these steps to start contributing: 61 | 62 | 1. Fork the [repository](https://github.com/lchen001/hapi) by 63 | clicking on the 'Fork' button on the repository's page. 64 | This creates a copy of the code under your GitHub user account. 65 | 66 | 2. Clone your fork to your local disk, and add the base repository as a remote: 67 | 68 | ```bash 69 | $ git clone git@github.com:/hapi.git 70 | $ cd hapi 71 | $ git remote add upstream https://github.com/lchen001/hapi.git 72 | ``` 73 | 74 | 3. Create a new branch to hold your development changes: 75 | 76 | ```bash 77 | $ git checkout -b a-descriptive-name-for-my-changes 78 | ``` 79 | 80 | **Do not** work on the `main` branch. 81 | 82 | 4. hapi manages dependencies using [`poetry`](https://python-poetry.org). 83 | Set up a development environment with `poetry` by running the following command in 84 | a virtual environment: 85 | 86 | ```bash 87 | $ pip install poetry 88 | $ poetry install 89 | ``` 90 | Note: in order to pass the full test suite (step 5), you'll need to install all extra in addition. 91 | ```bash 92 | $ poetry install --extras "adversarial augmentation summarization text vision" 93 | ``` 94 | 5. Develop features on your branch. 95 | 96 | As you work on the features, you should make sure that the test suite 97 | passes: 98 | 99 | ```bash 100 | $ pytest 101 | ``` 102 | 103 | hapi relies on `black` and `isort` to format its source code 104 | consistently. After you make changes, autoformat them with: 105 | 106 | ```bash 107 | $ make autoformat 108 | ``` 109 | 110 | hapi also uses `flake8` to check for coding mistakes. Quality control 111 | runs in CI, however you should also run the same checks with: 112 | 113 | ```bash 114 | $ make lint 115 | ``` 116 | 117 | If you're modifying documents under `docs/source`, make sure to validate that 118 | they can still be built. This check also runs in CI. To run a local check 119 | make sure you have installed the documentation builder requirements, by 120 | running `pip install -r docs/requirements.txt` from the root of this repository 121 | and then run: 122 | 123 | ```bash 124 | $ make docs 125 | ``` 126 | 127 | Once you're happy with your changes, add changed files using `git add` and 128 | make a commit with `git commit` to record your changes locally: 129 | 130 | ```bash 131 | $ git add modified_file.py 132 | $ git commit 133 | ``` 134 | 135 | Please write [good commit messages](https://chris.beams.io/posts/git-commit/). 136 | 137 | It is a good idea to sync your copy of the code with the original 138 | repository regularly. This way you can quickly account for changes: 139 | 140 | ```bash 141 | $ git fetch upstream 142 | $ git rebase upstream/main 143 | ``` 144 | 145 | Push the changes to your account using: 146 | 147 | ```bash 148 | $ git push -u origin a-descriptive-name-for-my-changes 149 | ``` 150 | 151 | You can use `pre-commit` to make sure you don't forget to format your code properly, 152 | the dependency should already be made available by `poetry`. 153 | 154 | Just install `pre-commit` for the `hapi` directory, 155 | 156 | ```bash 157 | $ pre-commit install 158 | ``` 159 | 160 | 6. Once you are satisfied (**and the checklist below is happy too**), go to the 161 | webpage of your fork on GitHub. Click on 'Pull request' to send your changes 162 | to the project maintainers for review. 163 | 164 | 7. It's ok if maintainers ask you for changes. It happens to core contributors 165 | too! So everyone can see the changes in the Pull request, work in your local 166 | branch and push the changes to your fork. They will automatically appear in 167 | the pull request. 168 | 169 | 8. We follow a one-commit-per-PR policy. Before your PR can be merged, you will have to 170 | `git rebase` to squash your changes into a single commit. 171 | 172 | ### Checklist 173 | 174 | 0. One commit per PR. 175 | 1. The title of your pull request should be a summary of its contribution; 176 | 2. If your pull request addresses an issue, please mention the issue number in 177 | the pull request description to make sure they are linked (and people 178 | consulting the issue know you are working on it); 179 | 3. To indicate a work in progress please prefix the title with `[WIP]`. These 180 | are useful to avoid duplicated work, and to differentiate it from PRs ready 181 | to be merged; 182 | 4. Make sure existing tests pass; 183 | 5. Add high-coverage tests. No quality testing = no merge. 184 | 6. All public methods must have informative docstrings that work nicely with sphinx. 185 | 186 | 187 | ### Tests 188 | 189 | A test suite is included to test the library behavior. 190 | Library tests can be found in the 191 | [tests folder](https://github.com/lchen001hapi/tree/main/tests). 192 | 193 | From the root of the 194 | repository, here's how to run tests with `pytest` for the library: 195 | 196 | ```bash 197 | $ make test 198 | ``` 199 | 200 | You can specify a smaller set of tests in order to test only the feature 201 | you're working on. 202 | 203 | Per the checklist above, all PRs should include high-coverage tests. 204 | To produce a code coverage report, run the following `pytest` 205 | ``` 206 | pytest --cov-report term-missing,html --cov=hapi . 207 | ``` 208 | This will populate a directory `htmlcov` with an HTML report. 209 | Open `htmlcov/index.html` in a browser to view the report. 210 | 211 | 212 | ### Style guide 213 | 214 | For documentation strings, hapi follows the 215 | [google style](https://google.github.io/styleguide/pyguide.html). -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /Makefile: -------------------------------------------------------------------------------- 1 | autoformat: 2 | black dcbench/ tests/ 3 | autoflake --in-place --remove-all-unused-imports -r dcbench tests 4 | isort --atomic dcbench/ tests/ 5 | docformatter --in-place --recursive dcbench tests 6 | 7 | lint: 8 | isort -c dcbench/ tests/ 9 | black dcbench/ tests/ --check 10 | flake8 dcbench/ tests/ 11 | 12 | test: 13 | pytest 14 | 15 | test-basic: 16 | set -e 17 | python -c "import dcbench as mk" 18 | python -c "import dcbench.version as mversion" 19 | 20 | test-cov: 21 | pytest --cov=./ --cov-report=xml 22 | 23 | docs: 24 | sphinx-build -b html docs/source/ docs/build/html/ 25 | 26 | docs-check: 27 | sphinx-build -b html docs/source/ docs/build/html/ -W 28 | 29 | livedocs: 30 | sphinx-autobuild -b html docs/source/ docs/build/html/ 31 | 32 | dev: 33 | pip install black isort flake8 docformatter pytest-cov sphinx-rtd-theme nbsphinx recommonmark pre-commit 34 | 35 | all: autoformat lint docs test 36 | -------------------------------------------------------------------------------- /Pipfile: -------------------------------------------------------------------------------- 1 | [[source]] 2 | url = "https://pypi.org/simple" 3 | verify_ssl = true 4 | name = "pypi" 5 | 6 | [dev-packages] 7 | twine = "*" 8 | ipython = "*" 9 | 10 | [requires] 11 | python_version = "3.8" 12 | 13 | [dev-packages.dcbench] 14 | path = "." 15 | extras = [ "dev",] 16 | editable = true 17 | 18 | [packages.dcbench] 19 | editable = true 20 | path = "." 21 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | 2 |
3 | banner 4 | 5 | ----- 6 | 7 | 8 | [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit&logoColor=white)](https://github.com/pre-commit/pre-commit) 9 | [![standard-readme compliant](https://img.shields.io/badge/readme%20style-standard-brightgreen.svg?style=flat-square)](https://github.com/RichardLitt/standard-readme) 10 | [![Paper](http://img.shields.io/badge/paper-arxiv.2209.08443-B31B1B.svg)](https://arxiv.org/abs/2209.08443) 11 | [![License](https://img.shields.io/badge/license-Apache%202-blue.svg)](LICENSE) 12 | [![Conference](http://img.shields.io/badge/NeurIPS-2022-4b44ce.svg)]() 13 | 14 | 15 | A longitudinal database of ML API predictions. 16 | 17 | [**Getting Started**](#%EF%B8%8F-quickstart) 18 | | [**Website**](http://hapi.stanford.edu/) 19 | | [**Contributing**](CONTRIBUTING.md) 20 | | [**About**](#%EF%B8%8F-about) 21 |
22 | 23 | 24 | ## 💡 What is HAPI? 25 | 26 |
27 | 28 | History of APIs (HAPI) is a large-scale, longitudinal database of commercial ML API predictions. It contains 1.7 million predictions collected from 2020 to 2022 and spanning APIs from Amazon, Google, IBM, and Microsoft. The database include diverse machine learning tasks including image tagging, speech recognition and text mining. 29 | 30 | 31 | 32 | ## ⚡️ Quickstart 33 | We provide a lightweight python package for getting started with HAPI. 34 | 35 | Read the guide below or follow along in Google Colab: 36 | 37 | [![Open intro](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/lchen001/HAPI/blob/main/examples/01_hapi_intro.ipynb) 38 | 39 | ```bash 40 | pip install "hapi @ git+https://github.com/lchen001/hapi@main" 41 | ``` 42 | 43 | Import the library and download the data, optionally specifying the directory for the 44 | the download. If the directory is not specified, the data will be downloaded to `~/.hapi`. 45 | 46 | 47 | ```python 48 | >> import hapi 49 | 50 | >> hapi.config.data_dir = "/path/to/data/dir" 51 | 52 | >> hapi.download() 53 | ``` 54 | 55 | > You can permanently set the data directory by adding the variable `HAPI_DATA_DIR` to your environment. 56 | 57 | Once we've downloaded the database, we can list the available APIs, datasets, and tasks with `hapi.summary()`. This returns a [Pandas DataFrame](https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.html) with columns `task, dataset, api, date, path, cost_per_10k`. 58 | ```python 59 | >> df = hapi.summary() 60 | ``` 61 | 62 | To load the predictions into memory we use `hapi.get_predictions()`. The keyword arguments allow us to load predictions for a subset of tasks, datasets, apis and/or dates. 63 | ```python 64 | >> predictions = hapi.get_predictions(task="mic", dataset="pascal", api=["google_mic", "ibm_mic"]) 65 | ``` 66 | 67 | The predictions are returned as a dictionary mapping from `"{task}/{dataset}/{api}/{date}"` to lists of dictionaries, each with keys `"example_id"`, `"predicted_label"`, and `"confidence"`. For example: 68 | ```python 69 | { 70 | "mic/pascal/google_mic/20-10-28": [ 71 | { 72 | 'confidence': 0.9798267782, 73 | 'example_id': '2011_000494', 74 | 'predicted_label': ['bird', 'bird'] 75 | }, 76 | ... 77 | ], 78 | "mic/pascal/microsoft_mic/20-10-28": [...], 79 | ... 80 | } 81 | ``` 82 | 83 | To load the labels into memory we use `hapi.get_labels()`. The keyword arguments allow us to load labels for a subset of tasks and datasets. 84 | ```python 85 | >> labels = hapi.get_labels(task="mic", dataset="pascal") 86 | ``` 87 | 88 | The labels are returned as a dictionary mapping from `"{task}/{dataset}"` to lists of dictionaries, each with keys `"example_id"` and `"true_label"`. 89 | 90 | 91 | ## 💾 Manual Downloading 92 | In this section, we discuss how to download the database without the HAPI Python API. 93 | 94 | The database is stored in a GCP bucket named [`hapi-data`](https://console.cloud.google.com/storage/browser/hapi-data). All model predictions are stored in [`hapi.tar.gz`](https://storage.googleapis.com/hapi-data/hapi.tar.gz) (Compressed size: `205.3MB`, Full size: `1.2GB`). 95 | 96 | From the command line, you can download and extract the predictions with: 97 | ```bash 98 | wget https://storage.googleapis.com/hapi-data/hapi.tar.gz && tar -xzvf hapi.tar.gz 99 | ``` 100 | However, we recommend downloading using the Python API as described above. 101 | 102 | 103 | ## 🌍 Datasets 104 | In this section, we discuss how to download the benchmark datasets used in HAPI. 105 | 106 | The predictions in HAPI are made on benchmark datasets from across the machine learning community. For example, HAPI includes predictions on [PASCAL](http://host.robots.ox.ac.uk/pascal/VOC/), a popular object detection dataset. Unfortunately, we lack the permissions required to redistribute these datasets, so we do not include the raw data in the download described above. 107 | 108 | We provide instructions on how to download each of the datasets and, for a growing number of them, we provide automated scripts that can download the dataset. These scripts are implemented in the [Meerkat Dataset Registry](https://meerkat.readthedocs.io/en/dev/datasets/datasets.html) – a registry of machine learning datasets (similar to [Torchvision Datasets](https://pytorch.org/vision/stable/datasets.html)). 109 | 110 | To download a dataset and load it into memory, use `hapi.get_dataset()`: 111 | ```python 112 | >> import hapi 113 | >> dp = hapi.get_dataset("pascal") 114 | ``` 115 | This returns a [Meerkat DataPanel](https://meerkat.readthedocs.io/en/latest/guide/data_structures.html#datapanel) – a DataFrame-like object that houses the dataset. See the Meerkat [User Guide](https://meerkat.readthedocs.io/en/latest/guide/guide.html) for more information. The DataPanel will have an "example_id" column that corresponds to the "example_id" key in the outputs of `hapi.get_predictions()` and `hapi.get_labels()`. 116 | 117 | If the dataset is not yet available through the Meerkat Dataset Registry, a `ValueError` will be raised containing instructions for manually downloading the dataset. For example: 118 | 119 | ```python 120 | >> dp = hapi.get_dataset("cmd") 121 | 122 | ValueError: Data download for 'cmd' not yet available for download through the HAPI Python API. Please download manually following the instructions below: 123 | 124 | CMD is a spoken command recognition dataset. 125 | 126 | It can be downloaded here: https://pyroomacoustics.readthedocs.io/en/pypi-release/pyroomacoustics.datasets.google_speech_commands.html. 127 | ``` 128 | 129 | ## ✉️ About 130 | `HAPI` was developed at Stanford in the Zou Group. Reach out to Lingjiao Chen (lingjiao [at] stanford [dot] edu) and Sabri Eyuboglu (eyuboglu [at] stanford [dot] edu) if you would like to get involved! 131 | -------------------------------------------------------------------------------- /docs/assets/banner.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lchen001/HAPI/dbf91fa7771e896d5dfe1a5d3e09a04fca2f65e1/docs/assets/banner.png -------------------------------------------------------------------------------- /docs/assets/datasets.b9bdb1e1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lchen001/HAPI/dbf91fa7771e896d5dfe1a5d3e09a04fca2f65e1/docs/assets/datasets.b9bdb1e1.png -------------------------------------------------------------------------------- /examples/01_hapi_intro.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "nbformat": 4, 3 | "nbformat_minor": 0, 4 | "metadata": { 5 | "colab": { 6 | "name": "try-HAPI.ipynb", 7 | "provenance": [], 8 | "collapsed_sections": [] 9 | }, 10 | "kernelspec": { 11 | "name": "python3", 12 | "display_name": "Python 3" 13 | }, 14 | "language_info": { 15 | "name": "python" 16 | }, 17 | "widgets": { 18 | "application/vnd.jupyter.widget-state+json": { 19 | "40e82a15b15f4de5afdb78193332d6d6": { 20 | "model_module": "@jupyter-widgets/controls", 21 | "model_name": "HBoxModel", 22 | "model_module_version": "1.5.0", 23 | "state": { 24 | "_dom_classes": [], 25 | "_model_module": "@jupyter-widgets/controls", 26 | "_model_module_version": "1.5.0", 27 | "_model_name": "HBoxModel", 28 | "_view_count": null, 29 | "_view_module": "@jupyter-widgets/controls", 30 | "_view_module_version": "1.5.0", 31 | "_view_name": "HBoxView", 32 | "box_style": "", 33 | "children": [ 34 | "IPY_MODEL_a0963e575f104ad280db8f98704aaaa7", 35 | "IPY_MODEL_9fa0f34d1a7c4f5e8a6f0e438b32bec0", 36 | "IPY_MODEL_3b66832981354857925b58f87e509734" 37 | ], 38 | "layout": "IPY_MODEL_ff2536158a1048ecaba0a6a0d14ba47a" 39 | } 40 | }, 41 | "a0963e575f104ad280db8f98704aaaa7": { 42 | "model_module": "@jupyter-widgets/controls", 43 | "model_name": "HTMLModel", 44 | "model_module_version": "1.5.0", 45 | "state": { 46 | "_dom_classes": [], 47 | "_model_module": "@jupyter-widgets/controls", 48 | "_model_module_version": "1.5.0", 49 | "_model_name": "HTMLModel", 50 | "_view_count": null, 51 | "_view_module": "@jupyter-widgets/controls", 52 | "_view_module_version": "1.5.0", 53 | "_view_name": "HTMLView", 54 | "description": "", 55 | "description_tooltip": null, 56 | "layout": "IPY_MODEL_e359003626b345068482112b93d09212", 57 | "placeholder": "​", 58 | "style": "IPY_MODEL_de72432d9c114c8d86e92eb3b771f305", 59 | "value": "100%" 60 | } 61 | }, 62 | "9fa0f34d1a7c4f5e8a6f0e438b32bec0": { 63 | "model_module": "@jupyter-widgets/controls", 64 | "model_name": "FloatProgressModel", 65 | "model_module_version": "1.5.0", 66 | "state": { 67 | "_dom_classes": [], 68 | "_model_module": "@jupyter-widgets/controls", 69 | "_model_module_version": "1.5.0", 70 | "_model_name": "FloatProgressModel", 71 | "_view_count": null, 72 | "_view_module": "@jupyter-widgets/controls", 73 | "_view_module_version": "1.5.0", 74 | "_view_name": "ProgressView", 75 | "bar_style": "success", 76 | "description": "", 77 | "description_tooltip": null, 78 | "layout": "IPY_MODEL_e0f153a65f6348bd846a9d27b1314db6", 79 | "max": 4, 80 | "min": 0, 81 | "orientation": "horizontal", 82 | "style": "IPY_MODEL_9d402dca42ab4e138bd6354c687443f3", 83 | "value": 4 84 | } 85 | }, 86 | "3b66832981354857925b58f87e509734": { 87 | "model_module": "@jupyter-widgets/controls", 88 | "model_name": "HTMLModel", 89 | "model_module_version": "1.5.0", 90 | "state": { 91 | "_dom_classes": [], 92 | "_model_module": "@jupyter-widgets/controls", 93 | "_model_module_version": "1.5.0", 94 | "_model_name": "HTMLModel", 95 | "_view_count": null, 96 | "_view_module": "@jupyter-widgets/controls", 97 | "_view_module_version": "1.5.0", 98 | "_view_name": "HTMLView", 99 | "description": "", 100 | "description_tooltip": null, 101 | "layout": "IPY_MODEL_173824c85c264b90837f16ec27fa0889", 102 | "placeholder": "​", 103 | "style": "IPY_MODEL_8d25f5cb44a24c759ada87eb3f3ae1b8", 104 | "value": " 4/4 [00:02<00:00, 1.52it/s]" 105 | } 106 | }, 107 | "ff2536158a1048ecaba0a6a0d14ba47a": { 108 | "model_module": "@jupyter-widgets/base", 109 | "model_name": "LayoutModel", 110 | "model_module_version": "1.2.0", 111 | "state": { 112 | "_model_module": "@jupyter-widgets/base", 113 | "_model_module_version": "1.2.0", 114 | "_model_name": "LayoutModel", 115 | "_view_count": null, 116 | "_view_module": "@jupyter-widgets/base", 117 | "_view_module_version": "1.2.0", 118 | "_view_name": "LayoutView", 119 | "align_content": null, 120 | "align_items": null, 121 | "align_self": null, 122 | "border": null, 123 | "bottom": null, 124 | "display": null, 125 | "flex": null, 126 | "flex_flow": null, 127 | "grid_area": null, 128 | "grid_auto_columns": null, 129 | "grid_auto_flow": null, 130 | "grid_auto_rows": null, 131 | "grid_column": null, 132 | "grid_gap": null, 133 | "grid_row": null, 134 | "grid_template_areas": null, 135 | "grid_template_columns": null, 136 | "grid_template_rows": null, 137 | "height": null, 138 | "justify_content": null, 139 | "justify_items": null, 140 | "left": null, 141 | "margin": null, 142 | "max_height": null, 143 | "max_width": null, 144 | "min_height": null, 145 | "min_width": null, 146 | "object_fit": null, 147 | "object_position": null, 148 | "order": null, 149 | "overflow": null, 150 | "overflow_x": null, 151 | "overflow_y": null, 152 | "padding": null, 153 | "right": null, 154 | "top": null, 155 | "visibility": null, 156 | "width": null 157 | } 158 | }, 159 | "e359003626b345068482112b93d09212": { 160 | "model_module": "@jupyter-widgets/base", 161 | "model_name": "LayoutModel", 162 | "model_module_version": "1.2.0", 163 | "state": { 164 | "_model_module": "@jupyter-widgets/base", 165 | "_model_module_version": "1.2.0", 166 | "_model_name": "LayoutModel", 167 | "_view_count": null, 168 | "_view_module": "@jupyter-widgets/base", 169 | "_view_module_version": "1.2.0", 170 | "_view_name": "LayoutView", 171 | "align_content": null, 172 | "align_items": null, 173 | "align_self": null, 174 | "border": null, 175 | "bottom": null, 176 | "display": null, 177 | "flex": null, 178 | "flex_flow": null, 179 | "grid_area": null, 180 | "grid_auto_columns": null, 181 | "grid_auto_flow": null, 182 | "grid_auto_rows": null, 183 | "grid_column": null, 184 | "grid_gap": null, 185 | "grid_row": null, 186 | "grid_template_areas": null, 187 | "grid_template_columns": null, 188 | "grid_template_rows": null, 189 | "height": null, 190 | "justify_content": null, 191 | "justify_items": null, 192 | "left": null, 193 | "margin": null, 194 | "max_height": null, 195 | "max_width": null, 196 | "min_height": null, 197 | "min_width": null, 198 | "object_fit": null, 199 | "object_position": null, 200 | "order": null, 201 | "overflow": null, 202 | "overflow_x": null, 203 | "overflow_y": null, 204 | "padding": null, 205 | "right": null, 206 | "top": null, 207 | "visibility": null, 208 | "width": null 209 | } 210 | }, 211 | "de72432d9c114c8d86e92eb3b771f305": { 212 | "model_module": "@jupyter-widgets/controls", 213 | "model_name": "DescriptionStyleModel", 214 | "model_module_version": "1.5.0", 215 | "state": { 216 | "_model_module": "@jupyter-widgets/controls", 217 | "_model_module_version": "1.5.0", 218 | "_model_name": "DescriptionStyleModel", 219 | "_view_count": null, 220 | "_view_module": "@jupyter-widgets/base", 221 | "_view_module_version": "1.2.0", 222 | "_view_name": "StyleView", 223 | "description_width": "" 224 | } 225 | }, 226 | "e0f153a65f6348bd846a9d27b1314db6": { 227 | "model_module": "@jupyter-widgets/base", 228 | "model_name": "LayoutModel", 229 | "model_module_version": "1.2.0", 230 | "state": { 231 | "_model_module": "@jupyter-widgets/base", 232 | "_model_module_version": "1.2.0", 233 | "_model_name": "LayoutModel", 234 | "_view_count": null, 235 | "_view_module": "@jupyter-widgets/base", 236 | "_view_module_version": "1.2.0", 237 | "_view_name": "LayoutView", 238 | "align_content": null, 239 | "align_items": null, 240 | "align_self": null, 241 | "border": null, 242 | "bottom": null, 243 | "display": null, 244 | "flex": null, 245 | "flex_flow": null, 246 | "grid_area": null, 247 | "grid_auto_columns": null, 248 | "grid_auto_flow": null, 249 | "grid_auto_rows": null, 250 | "grid_column": null, 251 | "grid_gap": null, 252 | "grid_row": null, 253 | "grid_template_areas": null, 254 | "grid_template_columns": null, 255 | "grid_template_rows": null, 256 | "height": null, 257 | "justify_content": null, 258 | "justify_items": null, 259 | "left": null, 260 | "margin": null, 261 | "max_height": null, 262 | "max_width": null, 263 | "min_height": null, 264 | "min_width": null, 265 | "object_fit": null, 266 | "object_position": null, 267 | "order": null, 268 | "overflow": null, 269 | "overflow_x": null, 270 | "overflow_y": null, 271 | "padding": null, 272 | "right": null, 273 | "top": null, 274 | "visibility": null, 275 | "width": null 276 | } 277 | }, 278 | "9d402dca42ab4e138bd6354c687443f3": { 279 | "model_module": "@jupyter-widgets/controls", 280 | "model_name": "ProgressStyleModel", 281 | "model_module_version": "1.5.0", 282 | "state": { 283 | "_model_module": "@jupyter-widgets/controls", 284 | "_model_module_version": "1.5.0", 285 | "_model_name": "ProgressStyleModel", 286 | "_view_count": null, 287 | "_view_module": "@jupyter-widgets/base", 288 | "_view_module_version": "1.2.0", 289 | "_view_name": "StyleView", 290 | "bar_color": null, 291 | "description_width": "" 292 | } 293 | }, 294 | "173824c85c264b90837f16ec27fa0889": { 295 | "model_module": "@jupyter-widgets/base", 296 | "model_name": "LayoutModel", 297 | "model_module_version": "1.2.0", 298 | "state": { 299 | "_model_module": "@jupyter-widgets/base", 300 | "_model_module_version": "1.2.0", 301 | "_model_name": "LayoutModel", 302 | "_view_count": null, 303 | "_view_module": "@jupyter-widgets/base", 304 | "_view_module_version": "1.2.0", 305 | "_view_name": "LayoutView", 306 | "align_content": null, 307 | "align_items": null, 308 | "align_self": null, 309 | "border": null, 310 | "bottom": null, 311 | "display": null, 312 | "flex": null, 313 | "flex_flow": null, 314 | "grid_area": null, 315 | "grid_auto_columns": null, 316 | "grid_auto_flow": null, 317 | "grid_auto_rows": null, 318 | "grid_column": null, 319 | "grid_gap": null, 320 | "grid_row": null, 321 | "grid_template_areas": null, 322 | "grid_template_columns": null, 323 | "grid_template_rows": null, 324 | "height": null, 325 | "justify_content": null, 326 | "justify_items": null, 327 | "left": null, 328 | "margin": null, 329 | "max_height": null, 330 | "max_width": null, 331 | "min_height": null, 332 | "min_width": null, 333 | "object_fit": null, 334 | "object_position": null, 335 | "order": null, 336 | "overflow": null, 337 | "overflow_x": null, 338 | "overflow_y": null, 339 | "padding": null, 340 | "right": null, 341 | "top": null, 342 | "visibility": null, 343 | "width": null 344 | } 345 | }, 346 | "8d25f5cb44a24c759ada87eb3f3ae1b8": { 347 | "model_module": "@jupyter-widgets/controls", 348 | "model_name": "DescriptionStyleModel", 349 | "model_module_version": "1.5.0", 350 | "state": { 351 | "_model_module": "@jupyter-widgets/controls", 352 | "_model_module_version": "1.5.0", 353 | "_model_name": "DescriptionStyleModel", 354 | "_view_count": null, 355 | "_view_module": "@jupyter-widgets/base", 356 | "_view_module_version": "1.2.0", 357 | "_view_name": "StyleView", 358 | "description_width": "" 359 | } 360 | }, 361 | "e3bbee1863bc4f169e4c509d0c20eb56": { 362 | "model_module": "@jupyter-widgets/controls", 363 | "model_name": "HBoxModel", 364 | "model_module_version": "1.5.0", 365 | "state": { 366 | "_dom_classes": [], 367 | "_model_module": "@jupyter-widgets/controls", 368 | "_model_module_version": "1.5.0", 369 | "_model_name": "HBoxModel", 370 | "_view_count": null, 371 | "_view_module": "@jupyter-widgets/controls", 372 | "_view_module_version": "1.5.0", 373 | "_view_name": "HBoxView", 374 | "box_style": "", 375 | "children": [ 376 | "IPY_MODEL_1762b9b0efd84effb037585d65b56396", 377 | "IPY_MODEL_9317f9d225324c41a069bedc3c19afd8", 378 | "IPY_MODEL_758b16ef3e6a4963b278205ff7ad940a" 379 | ], 380 | "layout": "IPY_MODEL_afa7f4cb07534a389a9566954fd85b06" 381 | } 382 | }, 383 | "1762b9b0efd84effb037585d65b56396": { 384 | "model_module": "@jupyter-widgets/controls", 385 | "model_name": "HTMLModel", 386 | "model_module_version": "1.5.0", 387 | "state": { 388 | "_dom_classes": [], 389 | "_model_module": "@jupyter-widgets/controls", 390 | "_model_module_version": "1.5.0", 391 | "_model_name": "HTMLModel", 392 | "_view_count": null, 393 | "_view_module": "@jupyter-widgets/controls", 394 | "_view_module_version": "1.5.0", 395 | "_view_name": "HTMLView", 396 | "description": "", 397 | "description_tooltip": null, 398 | "layout": "IPY_MODEL_a797eb02975d46af85212bd56b869983", 399 | "placeholder": "​", 400 | "style": "IPY_MODEL_3900af1d46724aa48414c6989246a65f", 401 | "value": "100%" 402 | } 403 | }, 404 | "9317f9d225324c41a069bedc3c19afd8": { 405 | "model_module": "@jupyter-widgets/controls", 406 | "model_name": "FloatProgressModel", 407 | "model_module_version": "1.5.0", 408 | "state": { 409 | "_dom_classes": [], 410 | "_model_module": "@jupyter-widgets/controls", 411 | "_model_module_version": "1.5.0", 412 | "_model_name": "FloatProgressModel", 413 | "_view_count": null, 414 | "_view_module": "@jupyter-widgets/controls", 415 | "_view_module_version": "1.5.0", 416 | "_view_name": "ProgressView", 417 | "bar_style": "success", 418 | "description": "", 419 | "description_tooltip": null, 420 | "layout": "IPY_MODEL_4b3c658455a94bb0a713da5ee54eeb80", 421 | "max": 1, 422 | "min": 0, 423 | "orientation": "horizontal", 424 | "style": "IPY_MODEL_ebe6f3d7f6a34feb8b4763d5a36de8b2", 425 | "value": 1 426 | } 427 | }, 428 | "758b16ef3e6a4963b278205ff7ad940a": { 429 | "model_module": "@jupyter-widgets/controls", 430 | "model_name": "HTMLModel", 431 | "model_module_version": "1.5.0", 432 | "state": { 433 | "_dom_classes": [], 434 | "_model_module": "@jupyter-widgets/controls", 435 | "_model_module_version": "1.5.0", 436 | "_model_name": "HTMLModel", 437 | "_view_count": null, 438 | "_view_module": "@jupyter-widgets/controls", 439 | "_view_module_version": "1.5.0", 440 | "_view_name": "HTMLView", 441 | "description": "", 442 | "description_tooltip": null, 443 | "layout": "IPY_MODEL_5259a4379a97456b8f26c15b6207fd79", 444 | "placeholder": "​", 445 | "style": "IPY_MODEL_98bd390eb7df48af9ce7b3b5818ac797", 446 | "value": " 1/1 [00:00<00:00, 2.39it/s]" 447 | } 448 | }, 449 | "afa7f4cb07534a389a9566954fd85b06": { 450 | "model_module": "@jupyter-widgets/base", 451 | "model_name": "LayoutModel", 452 | "model_module_version": "1.2.0", 453 | "state": { 454 | "_model_module": "@jupyter-widgets/base", 455 | "_model_module_version": "1.2.0", 456 | "_model_name": "LayoutModel", 457 | "_view_count": null, 458 | "_view_module": "@jupyter-widgets/base", 459 | "_view_module_version": "1.2.0", 460 | "_view_name": "LayoutView", 461 | "align_content": null, 462 | "align_items": null, 463 | "align_self": null, 464 | "border": null, 465 | "bottom": null, 466 | "display": null, 467 | "flex": null, 468 | "flex_flow": null, 469 | "grid_area": null, 470 | "grid_auto_columns": null, 471 | "grid_auto_flow": null, 472 | "grid_auto_rows": null, 473 | "grid_column": null, 474 | "grid_gap": null, 475 | "grid_row": null, 476 | "grid_template_areas": null, 477 | "grid_template_columns": null, 478 | "grid_template_rows": null, 479 | "height": null, 480 | "justify_content": null, 481 | "justify_items": null, 482 | "left": null, 483 | "margin": null, 484 | "max_height": null, 485 | "max_width": null, 486 | "min_height": null, 487 | "min_width": null, 488 | "object_fit": null, 489 | "object_position": null, 490 | "order": null, 491 | "overflow": null, 492 | "overflow_x": null, 493 | "overflow_y": null, 494 | "padding": null, 495 | "right": null, 496 | "top": null, 497 | "visibility": null, 498 | "width": null 499 | } 500 | }, 501 | "a797eb02975d46af85212bd56b869983": { 502 | "model_module": "@jupyter-widgets/base", 503 | "model_name": "LayoutModel", 504 | "model_module_version": "1.2.0", 505 | "state": { 506 | "_model_module": "@jupyter-widgets/base", 507 | "_model_module_version": "1.2.0", 508 | "_model_name": "LayoutModel", 509 | "_view_count": null, 510 | "_view_module": "@jupyter-widgets/base", 511 | "_view_module_version": "1.2.0", 512 | "_view_name": "LayoutView", 513 | "align_content": null, 514 | "align_items": null, 515 | "align_self": null, 516 | "border": null, 517 | "bottom": null, 518 | "display": null, 519 | "flex": null, 520 | "flex_flow": null, 521 | "grid_area": null, 522 | "grid_auto_columns": null, 523 | "grid_auto_flow": null, 524 | "grid_auto_rows": null, 525 | "grid_column": null, 526 | "grid_gap": null, 527 | "grid_row": null, 528 | "grid_template_areas": null, 529 | "grid_template_columns": null, 530 | "grid_template_rows": null, 531 | "height": null, 532 | "justify_content": null, 533 | "justify_items": null, 534 | "left": null, 535 | "margin": null, 536 | "max_height": null, 537 | "max_width": null, 538 | "min_height": null, 539 | "min_width": null, 540 | "object_fit": null, 541 | "object_position": null, 542 | "order": null, 543 | "overflow": null, 544 | "overflow_x": null, 545 | "overflow_y": null, 546 | "padding": null, 547 | "right": null, 548 | "top": null, 549 | "visibility": null, 550 | "width": null 551 | } 552 | }, 553 | "3900af1d46724aa48414c6989246a65f": { 554 | "model_module": "@jupyter-widgets/controls", 555 | "model_name": "DescriptionStyleModel", 556 | "model_module_version": "1.5.0", 557 | "state": { 558 | "_model_module": "@jupyter-widgets/controls", 559 | "_model_module_version": "1.5.0", 560 | "_model_name": "DescriptionStyleModel", 561 | "_view_count": null, 562 | "_view_module": "@jupyter-widgets/base", 563 | "_view_module_version": "1.2.0", 564 | "_view_name": "StyleView", 565 | "description_width": "" 566 | } 567 | }, 568 | "4b3c658455a94bb0a713da5ee54eeb80": { 569 | "model_module": "@jupyter-widgets/base", 570 | "model_name": "LayoutModel", 571 | "model_module_version": "1.2.0", 572 | "state": { 573 | "_model_module": "@jupyter-widgets/base", 574 | "_model_module_version": "1.2.0", 575 | "_model_name": "LayoutModel", 576 | "_view_count": null, 577 | "_view_module": "@jupyter-widgets/base", 578 | "_view_module_version": "1.2.0", 579 | "_view_name": "LayoutView", 580 | "align_content": null, 581 | "align_items": null, 582 | "align_self": null, 583 | "border": null, 584 | "bottom": null, 585 | "display": null, 586 | "flex": null, 587 | "flex_flow": null, 588 | "grid_area": null, 589 | "grid_auto_columns": null, 590 | "grid_auto_flow": null, 591 | "grid_auto_rows": null, 592 | "grid_column": null, 593 | "grid_gap": null, 594 | "grid_row": null, 595 | "grid_template_areas": null, 596 | "grid_template_columns": null, 597 | "grid_template_rows": null, 598 | "height": null, 599 | "justify_content": null, 600 | "justify_items": null, 601 | "left": null, 602 | "margin": null, 603 | "max_height": null, 604 | "max_width": null, 605 | "min_height": null, 606 | "min_width": null, 607 | "object_fit": null, 608 | "object_position": null, 609 | "order": null, 610 | "overflow": null, 611 | "overflow_x": null, 612 | "overflow_y": null, 613 | "padding": null, 614 | "right": null, 615 | "top": null, 616 | "visibility": null, 617 | "width": null 618 | } 619 | }, 620 | "ebe6f3d7f6a34feb8b4763d5a36de8b2": { 621 | "model_module": "@jupyter-widgets/controls", 622 | "model_name": "ProgressStyleModel", 623 | "model_module_version": "1.5.0", 624 | "state": { 625 | "_model_module": "@jupyter-widgets/controls", 626 | "_model_module_version": "1.5.0", 627 | "_model_name": "ProgressStyleModel", 628 | "_view_count": null, 629 | "_view_module": "@jupyter-widgets/base", 630 | "_view_module_version": "1.2.0", 631 | "_view_name": "StyleView", 632 | "bar_color": null, 633 | "description_width": "" 634 | } 635 | }, 636 | "5259a4379a97456b8f26c15b6207fd79": { 637 | "model_module": "@jupyter-widgets/base", 638 | "model_name": "LayoutModel", 639 | "model_module_version": "1.2.0", 640 | "state": { 641 | "_model_module": "@jupyter-widgets/base", 642 | "_model_module_version": "1.2.0", 643 | "_model_name": "LayoutModel", 644 | "_view_count": null, 645 | "_view_module": "@jupyter-widgets/base", 646 | "_view_module_version": "1.2.0", 647 | "_view_name": "LayoutView", 648 | "align_content": null, 649 | "align_items": null, 650 | "align_self": null, 651 | "border": null, 652 | "bottom": null, 653 | "display": null, 654 | "flex": null, 655 | "flex_flow": null, 656 | "grid_area": null, 657 | "grid_auto_columns": null, 658 | "grid_auto_flow": null, 659 | "grid_auto_rows": null, 660 | "grid_column": null, 661 | "grid_gap": null, 662 | "grid_row": null, 663 | "grid_template_areas": null, 664 | "grid_template_columns": null, 665 | "grid_template_rows": null, 666 | "height": null, 667 | "justify_content": null, 668 | "justify_items": null, 669 | "left": null, 670 | "margin": null, 671 | "max_height": null, 672 | "max_width": null, 673 | "min_height": null, 674 | "min_width": null, 675 | "object_fit": null, 676 | "object_position": null, 677 | "order": null, 678 | "overflow": null, 679 | "overflow_x": null, 680 | "overflow_y": null, 681 | "padding": null, 682 | "right": null, 683 | "top": null, 684 | "visibility": null, 685 | "width": null 686 | } 687 | }, 688 | "98bd390eb7df48af9ce7b3b5818ac797": { 689 | "model_module": "@jupyter-widgets/controls", 690 | "model_name": "DescriptionStyleModel", 691 | "model_module_version": "1.5.0", 692 | "state": { 693 | "_model_module": "@jupyter-widgets/controls", 694 | "_model_module_version": "1.5.0", 695 | "_model_name": "DescriptionStyleModel", 696 | "_view_count": null, 697 | "_view_module": "@jupyter-widgets/base", 698 | "_view_module_version": "1.2.0", 699 | "_view_name": "StyleView", 700 | "description_width": "" 701 | } 702 | } 703 | } 704 | } 705 | }, 706 | "cells": [ 707 | { 708 | "cell_type": "markdown", 709 | "source": [ 710 | "![banner.png](data:image/png;base64,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)\n", 711 | "\n", 712 | "History of APIs (HAPI) is a large-scale, longitudinal database of commercial ML API predictions. It contains 1.7 million predictions collected from 2020 to 2022 and spanning APIs from Amazon, Google, IBM, and Microsoft. The database include diverse machine learning tasks including image tagging, speech recognition and text mining.\n", 713 | "\n", 714 | "This notebook will demonstrate how to get started with the database. " 715 | ], 716 | "metadata": { 717 | "id": "pafhshHp5Eoq" 718 | } 719 | }, 720 | { 721 | "cell_type": "markdown", 722 | "source": [ 723 | "We provide a lightweight Python package for getting started with HAPI. Let's install it with pip: " 724 | ], 725 | "metadata": { 726 | "id": "2lCXGKl44rrH" 727 | } 728 | }, 729 | { 730 | "cell_type": "code", 731 | "execution_count": null, 732 | "metadata": { 733 | "id": "V_BsBipBb9lD" 734 | }, 735 | "outputs": [], 736 | "source": [ 737 | "!pip install \"hapi@git+https://github.com/lchen001/hapi@main\"" 738 | ] 739 | }, 740 | { 741 | "cell_type": "markdown", 742 | "source": [ 743 | "Import the library and download the data, optionally specifying the directory for the the download. \n", 744 | "\n", 745 | "If the directory is not specified, the data will be downloaded to `~/.hapi`.\n", 746 | "\n", 747 | "> You can permanently set the data directory by adding the variable `HAPI_DATA_DIR` to your environment. " 748 | ], 749 | "metadata": { 750 | "id": "zSTSHm-C5ySy" 751 | } 752 | }, 753 | { 754 | "cell_type": "code", 755 | "source": [ 756 | "import hapi\n", 757 | "hapi.config.data_dir = \".\" \n", 758 | "hapi.download();" 759 | ], 760 | "metadata": { 761 | "id": "TLepmJA3cFp_" 762 | }, 763 | "execution_count": 5, 764 | "outputs": [] 765 | }, 766 | { 767 | "cell_type": "markdown", 768 | "source": [ 769 | "Once we've downloaded the database, we can list the available APIs, datasets, and tasks with `hapi.summary()`. This returns a [Pandas DataFrame](https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.html) with columns `task, dataset, api, date, path, cost_per_10k`. " 770 | ], 771 | "metadata": { 772 | "id": "nVmIrTD-59hp" 773 | } 774 | }, 775 | { 776 | "cell_type": "code", 777 | "source": [ 778 | "hapi.summary()" 779 | ], 780 | "metadata": { 781 | "colab": { 782 | "base_uri": "https://localhost:8080/", 783 | "height": 655 784 | }, 785 | "id": "wW17vea2cs-F", 786 | "outputId": "1fc99b1e-84b3-4f38-f140-a2820f0704dd" 787 | }, 788 | "execution_count": 6, 789 | "outputs": [ 790 | { 791 | "output_type": "execute_result", 792 | "data": { 793 | "text/plain": [ 794 | " task dataset api date \\\n", 795 | "0 scr command google_scr 20-03-29 \n", 796 | "1 scr command ibm_scr 20-03-29 \n", 797 | "2 scr command deepspeech_lib_scr 20-03-29 \n", 798 | "3 scr command microsoft_scr 20-03-29 \n", 799 | "4 scr command ibm_scr 22-05-23 \n", 800 | ".. ... ... ... ... \n", 801 | "171 fer ferplus facepp_fer 22-05-23 \n", 802 | "172 fer ferplus google_fer 22-05-23 \n", 803 | "173 sa imdb baidu_sa 21-02-21 \n", 804 | "174 sa imdb amazon_sa 21-02-21 \n", 805 | "175 sa imdb google_sa 21-02-21 \n", 806 | "\n", 807 | " path cost_per_10k \n", 808 | "0 scr/command/google_scr/20-03-29.json 60.00 \n", 809 | "1 scr/command/ibm_scr/20-03-29.json 25.00 \n", 810 | "2 scr/command/deepspeech_lib_scr/20-03-29.json 0.02 \n", 811 | "3 scr/command/microsoft_scr/20-03-29.json 41.00 \n", 812 | "4 scr/command/ibm_scr/22-05-23.json 25.00 \n", 813 | ".. ... ... \n", 814 | "171 fer/ferplus/facepp_fer/22-05-23.json 5.00 \n", 815 | "172 fer/ferplus/google_fer/22-05-23.json 15.00 \n", 816 | "173 sa/imdb/baidu_sa/21-02-21.json 3.50 \n", 817 | "174 sa/imdb/amazon_sa/21-02-21.json 0.75 \n", 818 | "175 sa/imdb/google_sa/21-02-21.json 2.50 \n", 819 | "\n", 820 | "[176 rows x 6 columns]" 821 | ], 822 | "text/html": [ 823 | "\n", 824 | "
\n", 825 | "
\n", 826 | "
\n", 827 | "\n", 840 | "\n", 841 | " \n", 842 | " \n", 843 | " \n", 844 | " \n", 845 | " \n", 846 | " \n", 847 | " \n", 848 | " \n", 849 | " \n", 850 | " \n", 851 | " \n", 852 | " \n", 853 | " \n", 854 | " \n", 855 | " \n", 856 | " \n", 857 | " \n", 858 | " \n", 859 | " \n", 860 | " \n", 861 | " \n", 862 | " \n", 863 | " \n", 864 | " \n", 865 | " \n", 866 | " \n", 867 | " \n", 868 | " \n", 869 | " \n", 870 | " \n", 871 | " \n", 872 | " \n", 873 | " \n", 874 | " \n", 875 | " \n", 876 | " \n", 877 | " \n", 878 | " \n", 879 | " \n", 880 | " \n", 881 | " \n", 882 | " \n", 883 | " \n", 884 | " \n", 885 | " \n", 886 | " \n", 887 | " \n", 888 | " \n", 889 | " \n", 890 | " \n", 891 | " \n", 892 | " \n", 893 | " \n", 894 | " \n", 895 | " \n", 896 | " \n", 897 | " \n", 898 | " \n", 899 | " \n", 900 | " \n", 901 | " \n", 902 | " \n", 903 | " \n", 904 | " \n", 905 | " \n", 906 | " \n", 907 | " \n", 908 | " \n", 909 | " \n", 910 | " \n", 911 | " \n", 912 | " \n", 913 | " \n", 914 | " \n", 915 | " \n", 916 | " \n", 917 | " \n", 918 | " \n", 919 | " \n", 920 | " \n", 921 | " \n", 922 | " \n", 923 | " \n", 924 | " \n", 925 | " \n", 926 | " \n", 927 | " \n", 928 | " \n", 929 | " \n", 930 | " \n", 931 | " \n", 932 | " \n", 933 | " \n", 934 | " \n", 935 | " \n", 936 | " \n", 937 | " \n", 938 | " \n", 939 | " \n", 940 | " \n", 941 | " \n", 942 | " \n", 943 | " \n", 944 | " \n", 945 | " \n", 946 | " \n", 947 | " \n", 948 | " \n", 949 | " \n", 950 | " \n", 951 | " \n", 952 | " \n", 953 | "
taskdatasetapidatepathcost_per_10k
0scrcommandgoogle_scr20-03-29scr/command/google_scr/20-03-29.json60.00
1scrcommandibm_scr20-03-29scr/command/ibm_scr/20-03-29.json25.00
2scrcommanddeepspeech_lib_scr20-03-29scr/command/deepspeech_lib_scr/20-03-29.json0.02
3scrcommandmicrosoft_scr20-03-29scr/command/microsoft_scr/20-03-29.json41.00
4scrcommandibm_scr22-05-23scr/command/ibm_scr/22-05-23.json25.00
.....................
171ferferplusfacepp_fer22-05-23fer/ferplus/facepp_fer/22-05-23.json5.00
172ferferplusgoogle_fer22-05-23fer/ferplus/google_fer/22-05-23.json15.00
173saimdbbaidu_sa21-02-21sa/imdb/baidu_sa/21-02-21.json3.50
174saimdbamazon_sa21-02-21sa/imdb/amazon_sa/21-02-21.json0.75
175saimdbgoogle_sa21-02-21sa/imdb/google_sa/21-02-21.json2.50
\n", 954 | "

176 rows × 6 columns

\n", 955 | "
\n", 956 | " \n", 966 | " \n", 967 | " \n", 1004 | "\n", 1005 | " \n", 1029 | "
\n", 1030 | "
\n", 1031 | " " 1032 | ] 1033 | }, 1034 | "metadata": {}, 1035 | "execution_count": 6 1036 | } 1037 | ] 1038 | }, 1039 | { 1040 | "cell_type": "markdown", 1041 | "source": [ 1042 | "To load the predictions into memory we use `hapi.get_predictions()`. The keyword arguments allow us to load predictions for a subset of tasks, datasets, apis and/or dates. \n", 1043 | "\n", 1044 | "The predictions are returned as a dictionary mapping from `\"{task}/{dataset}/{api}/{date}\"` to lists of dictionaries, each with keys `\"example_id\"`, `\"predicted_label\"`, and `\"confidence\"`." 1045 | ], 1046 | "metadata": { 1047 | "id": "OTsuXC5n6Tyf" 1048 | } 1049 | }, 1050 | { 1051 | "cell_type": "code", 1052 | "source": [ 1053 | "predictions = hapi.get_predictions(task=\"mic\", dataset=\"coco\", api=[\"google_mic\", \"microsoft_mic\"])\n", 1054 | "\n", 1055 | "predictions[\"mic/coco/microsoft_mic/20-11-20\"][:3]" 1056 | ], 1057 | "metadata": { 1058 | "colab": { 1059 | "base_uri": "https://localhost:8080/", 1060 | "height": 210, 1061 | "referenced_widgets": [ 1062 | "40e82a15b15f4de5afdb78193332d6d6", 1063 | "a0963e575f104ad280db8f98704aaaa7", 1064 | "9fa0f34d1a7c4f5e8a6f0e438b32bec0", 1065 | "3b66832981354857925b58f87e509734", 1066 | "ff2536158a1048ecaba0a6a0d14ba47a", 1067 | "e359003626b345068482112b93d09212", 1068 | "de72432d9c114c8d86e92eb3b771f305", 1069 | "e0f153a65f6348bd846a9d27b1314db6", 1070 | "9d402dca42ab4e138bd6354c687443f3", 1071 | "173824c85c264b90837f16ec27fa0889", 1072 | "8d25f5cb44a24c759ada87eb3f3ae1b8" 1073 | ] 1074 | }, 1075 | "id": "Ip2kbMGucwkU", 1076 | "outputId": "4f0a8cf7-d605-4250-f56b-aaa106e2087e" 1077 | }, 1078 | "execution_count": 21, 1079 | "outputs": [ 1080 | { 1081 | "output_type": "display_data", 1082 | "data": { 1083 | "text/plain": [ 1084 | " 0%| | 0/4 [00:00 str: 55 | """Download the HAPI database. 56 | 57 | The database is stored in a GCP bucket named hapi-data. All model predictions are 58 | stored in hapi.tar.gz (Compressed size: 205.3MB, Full size: 1.2GB). This function 59 | downloads the archive and extracts it. 60 | 61 | Args: 62 | data_dir (str, optional): Directory to download. Defaults to None, in which case 63 | `config.data_dir` is used. If `config.data_dir` is not set, then the default 64 | directory is used: `~/.hapi`. 65 | 66 | Returns: 67 | str: The path to the downloaded data. 68 | """ 69 | if data_dir is None: 70 | data_dir = config._data_dir 71 | 72 | os.makedirs(data_dir, exist_ok=True) 73 | 74 | urlretrieve( 75 | DATA_URL, 76 | os.path.join(data_dir, "hapi.tar.gz"), 77 | ) 78 | 79 | # extract the tarball 80 | import tarfile 81 | 82 | with tarfile.open(os.path.join(data_dir, "hapi.tar.gz")) as tar: 83 | tar.extractall(data_dir) 84 | 85 | return data_dir 86 | 87 | 88 | def get_predictions( 89 | task: Union[str, List[str]] = None, 90 | dataset: Union[str, List[str]] = None, 91 | api: Union[str, List[str]] = None, 92 | date: Union[str, List[str]] = None, 93 | include_dataset: bool = None, 94 | ) -> Dict[str, List[Dict]]: 95 | """Load API predictions into memory. 96 | 97 | Use the `task`, `dataset`, `api`, and `date` parameters to filter to a subset of 98 | the database. If more than one of these parameters is specified, the results will 99 | be filtered to include only those rows that match all of the specified filters 100 | (i.e. we apply AND logic). 101 | 102 | Args: 103 | task (Union[str, List[str]]): The task(s) to include. If None, all tasks are 104 | loaded. Default is None. Use ``hapi.summary()["task"].unique()`` to see 105 | options. 106 | dataset (Union[str, List[str]]): The dataset(s) to include. If None, all 107 | datasets are loaded. Default is None. Use 108 | ``hapi.summary()["dataset"].unique()`` to see options. 109 | api (Union[str, List[str]]): The API(s) to include. If None, all APIs are 110 | loaded. Default is None. Use ``hapi.summary()["api"].unique()`` to see 111 | options. 112 | date (Union[str, List[str]]): The date(s) to include in format "y-m-d". For 113 | example, "20-03-29". If None, all dates are loaded. Default is None. 114 | include_dataset (bool, optional): If True, the raw dataset is downloaded and 115 | loaded using `hapi.get_dataset()`. The dataset is then merged with the 116 | predictions on the "example_id" column. Default is False. 117 | 118 | Returns: 119 | Dict[str, List[Dict]]: A dictionary mapping keys in the format 120 | "{task}/{dataset}/{api)/{date}" (e.g. "scr/command/google_scr/20-03-29") to a 121 | list of dictionaries, each representing one prediction. These dictionaries 122 | include keys "confidence", "predicted_label", and "example_id". For example, 123 | 124 | .. code-block:: python 125 | 126 | { 127 | "scr/command/google_scr/20-03-29": [ 128 | { 129 | 'confidence': 0.9128385782, 130 | 'predicted_label': 0, 131 | 'example_id': 'COMMAND_004ae714_nohash_0.wav' 132 | }, 133 | ], 134 | ... 135 | } 136 | """ 137 | df = summary() 138 | if task is not None: 139 | if isinstance(task, str): 140 | df = df[df["task"] == task] 141 | else: 142 | df = df[df["task"].isin(task)] 143 | 144 | if dataset is not None: 145 | if isinstance(dataset, str): 146 | df = df[df["dataset"] == dataset] 147 | else: 148 | df = df[df["dataset"].isin(dataset)] 149 | 150 | if api is not None: 151 | if isinstance(api, str): 152 | df = df[df["api"] == api] 153 | else: 154 | df = df[df["api"].isin(api)] 155 | 156 | if date is not None: 157 | if isinstance(date, str): 158 | df = df[df["date"] == date] 159 | else: 160 | df = df[df["date"].isin(date)] 161 | 162 | if include_dataset: 163 | dataset_to_data = { 164 | dataset: get_dataset(dataset) 165 | for dataset in ([dataset] if isinstance(dataset, str) else dataset) 166 | } 167 | 168 | path_to_preds = {} 169 | for _, row in tqdm(df.iterrows(), total=len(df)): 170 | path = row["path"] 171 | preds = json.load(open(os.path.join(config.data_dir, "tasks", path))) 172 | 173 | if include_dataset: 174 | import meerkat as mk 175 | 176 | preds = mk.DataPanel(preds).merge( 177 | dataset_to_data[row["dataset"]], on="example_id" 178 | ) 179 | 180 | path_to_preds[os.path.splitext(path)[0]] = preds 181 | 182 | return path_to_preds 183 | 184 | 185 | def get_labels( 186 | task: Union[str, List[str]] = None, 187 | dataset: Union[str, List[str]] = None, 188 | ) -> Dict[str, List[Dict]]: 189 | """Load labels into memory. 190 | 191 | Use the `task` and `dataset` parameters to filter to a subset of the database. 192 | If more than one of these parameters is specified, the results will be filtered 193 | to include only those rows that match all of the specified filters (i.e. we apply 194 | AND logic). 195 | 196 | Args: 197 | task (Union[str, List[str]]): The task(s) to include. If None, all tasks are 198 | loaded. Default is None. Use ``hapi.summary()["task"].unique()`` to see 199 | options. 200 | dataset (Union[str, List[str]]): The dataset(s) to include. If None, all 201 | datasets are loaded. Default is None. Use 202 | ``hapi.summary()["dataset"].unique()`` to see options. 203 | 204 | Returns: 205 | Dict[str, List[Dict]]: A dictionary mapping keys in the format 206 | "{task}/{dataset}" (e.g. "scr/command") to a 207 | list of dictionaries, each representing one label. These dictionaries 208 | include keys "label", "example_id", and "confidence". For example, 209 | 210 | .. code-block:: python 211 | 212 | { 213 | "scr/command": [ 214 | { 215 | 'true_label': 0, 216 | 'example_id': 'COMMAND_004ae714_nohash_0.wav', 217 | }, 218 | ], 219 | ... 220 | } 221 | """ 222 | df = summary() 223 | df = df[["task", "dataset"]].drop_duplicates() 224 | if task is not None: 225 | if isinstance(task, str): 226 | df = df[df["task"] == task] 227 | else: 228 | df = df[df["task"].isin(task)] 229 | 230 | if dataset is not None: 231 | if isinstance(dataset, str): 232 | df = df[df["dataset"] == dataset] 233 | else: 234 | df = df[df["dataset"].isin(dataset)] 235 | 236 | path_to_labels = {} 237 | for _, row in tqdm(df.iterrows(), total=len(df)): 238 | path = os.path.join(row["task"], row["dataset"]) 239 | labels = json.load( 240 | open(os.path.join(config.data_dir, "tasks", path, "labels.json")) 241 | ) 242 | path_to_labels[path] = labels 243 | return path_to_labels 244 | 245 | 246 | def summary() -> pd.DataFrame: 247 | """Summarize the HAPI database. 248 | 249 | Returns: 250 | pd.DataFrame: A dataframe where each row corresponds to one instance of API 251 | predictions (i.e. predictions from a single api on a single dataset on a 252 | single date). The dataframe contains the following columns: "task", 253 | "dataset", "api", "date", "path", and "cost_per_10k". 254 | """ 255 | df = pd.read_csv(os.path.join(config.data_dir, "tasks", "meta.csv")) 256 | return df 257 | -------------------------------------------------------------------------------- /hapi/convert.py: -------------------------------------------------------------------------------- 1 | import os 2 | import tarfile 3 | import pandas as pd 4 | import json 5 | import meerkat as mk 6 | import re 7 | from google.cloud import storage 8 | from pyrsistent import l 9 | from tqdm import tqdm 10 | 11 | DATASET_TO_TASK = { 12 | "coco": "mic", 13 | "mir": "mic", 14 | "pascal": "mic", 15 | "gmb": "ner", 16 | "conll": "ner", 17 | "zhner": "ner", 18 | "lsvt": "str", 19 | "rects": "str", 20 | "mtwi": "str", 21 | "expw": "fer", 22 | "ferplus": "fer", 23 | "rafdb": "fer", 24 | "afnet": "fer", 25 | "imdb": "sa", 26 | "waimai": "sa", 27 | "yelp": "sa", 28 | "shop": "sa", 29 | "digit": "scr", 30 | "command": "scr", 31 | "amnist": "scr", 32 | "fluent": "scr", 33 | } 34 | 35 | 36 | DATA_DIR = "/Users/eyubogln/code/hapi/data/legacy" 37 | 38 | DST_DIR = "/Users/eyubogln/code/hapi/data/tasks" 39 | 40 | BUCKET_NAME = "hapi-data" 41 | 42 | 43 | def get_structured_predictions( 44 | predictions_dir: str, model: str, include_original: bool = False 45 | ): 46 | 47 | # regex pattern for converting from came 48 | pattern = re.compile(r"(? "mk.DataPanel": 35 | """ Load a dataset from the Meerkat registry. If the dataset is not yet downloaded, 36 | it will be downloaded automatically. Not all datasets in HAPI are supported: 37 | if the dataset is not yet available through the Meerkat Dataset Registry, a ` 38 | ValueError` will be raised containing instructions for manually downloading the 39 | dataset. For example: 40 | 41 | .. code-block:: python 42 | 43 | >> dp = hapi.get_dataset("cmd") 44 | 45 | ValueError: Data download for 'cmd' not yet available for download through the HAPI Python API. Please download manually following the instructions below: 46 | 47 | CMD is a spoken command recognition dataset. 48 | 49 | It can be downloaded here: https://pyroomacoustics.readthedocs.io/en/... 50 | 51 | Args: 52 | dataset (str): The name of the dataset. 53 | 54 | Raises: 55 | ValueError: If the dataset is not yet included in the registry. The ValueError 56 | will contain instructions for manually downloading the dataset. 57 | 58 | Returns: 59 | mk.DataPanel: A Meerkat DataPanel holding the dataset. A Meerkat DataPanel is a 60 | DataFrame-like object that houses the dataset. See the Meerkat User Guide 61 | for more information. The DataPanel will have an "example_id" column that 62 | corresponds to the "example_id" key in the outputs of 63 | `hapi.get_predictions()` and `hapi.get_labels()`. 64 | """ 65 | 66 | import meerkat as mk 67 | 68 | if dataset == "expw": 69 | dp = mk.get("expw") 70 | 71 | # remove file extension and add the face_id 72 | dp["example_id"] = ( 73 | dp["image_name"].str.replace(".jpg", "", regex=False) 74 | + "_" 75 | + dp["face_id_in_image"].astype(str) 76 | ) 77 | 78 | return dp 79 | 80 | elif dataset == "pascal": 81 | dp = mk.get("pascal") 82 | dp["example_id"] = dp["id"] 83 | dp.remove_column("id") 84 | return dp 85 | 86 | elif dataset == "coco": 87 | dp = mk.get("coco") 88 | dp["example_id"] = dp["coco_url"].apply( 89 | lambda x: os.path.splitext(os.path.basename(x))[0] 90 | ) 91 | dp.remove_column("id") 92 | return dp 93 | 94 | elif dataset == "mir": 95 | dp = mk.get("mirflickr") 96 | dp["example_id"] = dp["id"] 97 | dp.remove_column("id") 98 | return dp 99 | 100 | elif dataset in DATASET_INFO: 101 | raise ValueError( 102 | f"Data download for '{dataset}' not yet available for download through the " 103 | " HAPI Python API. Please download manually following the instructions " 104 | "below: \n \n" 105 | f"{DATASET_INFO[dataset]}" 106 | ) 107 | else: 108 | raise ValueError( 109 | f"Unknown dataset '{dataset}'. Please pass one of the following: " 110 | f"{list(DATASET_INFO.keys())}" 111 | ) 112 | -------------------------------------------------------------------------------- /hapi/version.py: -------------------------------------------------------------------------------- 1 | __version__ = "0.0.1" 2 | -------------------------------------------------------------------------------- /pyproject.toml: -------------------------------------------------------------------------------- 1 | [build-system] 2 | requires = ["setuptools", "wheel"] 3 | build-backend = "setuptools.build_meta:__legacy__" -------------------------------------------------------------------------------- /setup.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | 4 | # Note: To use the 'upload' functionality of this file, you must: 5 | # $ pipenv install twine --dev 6 | 7 | import io 8 | import os 9 | import sys 10 | from distutils.util import convert_path 11 | from shutil import rmtree 12 | 13 | from setuptools import Command, find_packages, setup 14 | 15 | main_ns = {} 16 | ver_path = convert_path("hapi/version.py") 17 | with open(ver_path) as ver_file: 18 | exec(ver_file.read(), main_ns) 19 | 20 | 21 | # Package meta-data. 22 | NAME = "hapi" 23 | DESCRIPTION = ( 24 | "This is a benchmark that tests various data-centric aspects of improving the " 25 | "quality of machine learning workflows." 26 | ) 27 | URL = "" 28 | EMAIL = "eyuboglu@stanford.edu" 29 | AUTHOR = "https://github.com/lchen001/HAPI" 30 | REQUIRES_PYTHON = ">=3.7.0" 31 | VERSION = main_ns["__version__"] 32 | 33 | REQUIRED = [ 34 | "pandas", 35 | "numpy>=1.18.0", 36 | "jsonlines>=1.2.0", 37 | "meerkat-ml[dev,vision,ml]", 38 | ] 39 | EXTRAS = { 40 | "dev": [ 41 | "black==21.5b0", 42 | "isort>=5.7.0", 43 | "autoflake", 44 | "flake8>=3.8.4", 45 | "mypy>=0.9", 46 | "docformatter>=1.4", 47 | "pytest-cov>=2.10.1", 48 | "sphinx-rtd-theme>=0.5.1", 49 | "nbsphinx>=0.8.0", 50 | "recommonmark>=0.7.1", 51 | "parameterized", 52 | "pre-commit>=2.9.3", 53 | "sphinx-autobuild", 54 | "google-cloud-storage", 55 | "furo", 56 | ], 57 | } 58 | 59 | # The rest you shouldn't have to touch too much :) 60 | # ------------------------------------------------ 61 | # Except, perhaps the License and Trove Classifiers! 62 | # If you do change the License, remember to change the Trove Classifier for that! 63 | 64 | here = os.path.abspath(os.path.dirname(__file__)) 65 | 66 | # Import the README and use it as the long-description. 67 | # Note: this will only work if 'README.md' is present in your MANIFEST.in file! 68 | try: 69 | with io.open(os.path.join(here, "README.md"), encoding="utf-8") as f: 70 | long_description = "\n" + f.read() 71 | except FileNotFoundError: 72 | long_description = DESCRIPTION 73 | 74 | # Load the package's __version__.py module as a dictionary. 75 | about = {} 76 | if not VERSION: 77 | project_slug = NAME.lower().replace("-", "_").replace(" ", "_") 78 | with open(os.path.join(here, project_slug, "__version__.py")) as f: 79 | exec(f.read(), about) 80 | else: 81 | about["__version__"] = VERSION 82 | 83 | 84 | class UploadCommand(Command): 85 | """Support setup.py upload.""" 86 | 87 | description = "Build and publish the package." 88 | user_options = [] 89 | 90 | @staticmethod 91 | def status(s): 92 | """Prints things in bold.""" 93 | print("\033[1m{0}\033[0m".format(s)) 94 | 95 | def initialize_options(self): 96 | pass 97 | 98 | def finalize_options(self): 99 | pass 100 | 101 | def run(self): 102 | try: 103 | self.status("Removing previous builds…") 104 | rmtree(os.path.join(here, "dist")) 105 | except OSError: 106 | pass 107 | 108 | self.status("Building Source and Wheel (universal) distribution…") 109 | os.system("{0} setup.py sdist bdist_wheel --universal".format(sys.executable)) 110 | 111 | self.status("Uploading the package to PyPI via Twine…") 112 | os.system("twine upload dist/*") 113 | 114 | self.status("Pushing git tags…") 115 | os.system("git tag v{0}".format(about["__version__"])) 116 | os.system("git push --tags") 117 | 118 | sys.exit() 119 | 120 | 121 | # Where the magic happens: 122 | setup( 123 | name=NAME, 124 | version=about["__version__"], 125 | description=DESCRIPTION, 126 | long_description=long_description, 127 | long_description_content_type="text/markdown", 128 | author=AUTHOR, 129 | author_email=EMAIL, 130 | python_requires=REQUIRES_PYTHON, 131 | url=URL, 132 | packages=find_packages(exclude=["tests", "*.tests", "*.tests.*", "tests.*"]), 133 | # If your package is a single module, use this instead of 'packages': 134 | # py_modules=['mypackage'], 135 | # entry_points={ 136 | # 'console_scripts': ['mycli=mymodule:cli'], 137 | # }, 138 | install_requires=REQUIRED, 139 | extras_require=EXTRAS, 140 | include_package_data=True, 141 | license="Apache 2.0", 142 | classifiers=[ 143 | # Trove classifiers 144 | # Full list: https://pypi.python.org/pypi?%3Aaction=list_classifiers 145 | "Programming Language :: Python", 146 | "Programming Language :: Python :: 3", 147 | "Programming Language :: Python :: 3.7", 148 | "Programming Language :: Python :: 3.8", 149 | "Programming Language :: Python :: 3.9", 150 | "Topic :: Scientific/Engineering :: Artificial Intelligence", 151 | ], 152 | # $ setup.py publish support. 153 | cmdclass={ 154 | "upload": UploadCommand, 155 | }, 156 | ) 157 | --------------------------------------------------------------------------------