├── .github └── workflows │ ├── python-publish.yml │ └── stalebot ├── .gitignore ├── .pre-commit-config.yaml ├── CITATION.cff ├── LICENSE ├── README.md ├── pyproject.toml ├── src └── ai_models │ ├── __init__.py │ ├── __main__.py │ ├── checkpoint.py │ ├── inputs │ ├── __init__.py │ ├── base.py │ ├── cds.py │ ├── compute.py │ ├── file.py │ ├── interpolate.py │ ├── mars.py │ ├── opendata.py │ ├── recenter.py │ └── transform.py │ ├── model.py │ ├── outputs │ └── __init__.py │ ├── remote │ ├── __init__.py │ ├── api.py │ ├── config.py │ └── model.py │ └── stepper.py └── tests ├── requirements.txt └── test_code.py /.github/workflows/python-publish.yml: -------------------------------------------------------------------------------- 1 | # This workflow will upload a Python Package using Twine when a release is created 2 | # For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries 3 | 4 | name: Upload Python Package 5 | 6 | on: 7 | 8 | push: {} 9 | 10 | release: 11 | types: [created] 12 | 13 | jobs: 14 | quality: 15 | name: Code QA 16 | runs-on: ubuntu-latest 17 | steps: 18 | # - run: sudo apt-get install -y pandoc # Needed by sphinx for notebooks 19 | - uses: actions/checkout@v4 20 | - uses: actions/setup-python@v5 21 | with: 22 | python-version: 3.x 23 | - uses: pre-commit/action@v3.0.1 24 | env: 25 | SKIP: no-commit-to-branch 26 | 27 | 28 | checks: 29 | strategy: 30 | fail-fast: false 31 | matrix: 32 | platform: ["ubuntu-latest", "macos-latest"] 33 | python-version: ["3.10"] 34 | 35 | name: Python ${{ matrix.python-version }} on ${{ matrix.platform }} 36 | runs-on: ${{ matrix.platform }} 37 | 38 | steps: 39 | - uses: actions/checkout@v4 40 | 41 | - uses: actions/setup-python@v2 42 | with: 43 | python-version: ${{ matrix.python-version }} 44 | 45 | - name: Install 46 | run: | 47 | python -m pip install --upgrade pip 48 | pip install pytest 49 | pip install -e . 50 | pip install -r tests/requirements.txt 51 | pip freeze 52 | 53 | - name: Tests 54 | run: pytest 55 | 56 | deploy: 57 | 58 | if: ${{ github.event_name == 'release' }} 59 | runs-on: ubuntu-latest 60 | needs: [checks, quality] 61 | 62 | steps: 63 | - uses: actions/checkout@v4 64 | 65 | - name: Set up Python 66 | uses: actions/setup-python@v2 67 | with: 68 | python-version: 3.x 69 | 70 | - name: Install dependencies 71 | run: | 72 | python -m pip install --upgrade pip 73 | pip install build wheel twine 74 | - name: Build and publish 75 | env: 76 | TWINE_USERNAME: __token__ 77 | TWINE_PASSWORD: ${{ secrets.PYPI_API_TOKEN }} 78 | run: | 79 | python -m build 80 | twine upload dist/* 81 | -------------------------------------------------------------------------------- /.github/workflows/stalebot: -------------------------------------------------------------------------------- 1 | name: 'Close stale issues and PR' 2 | on: 3 | schedule: 4 | - cron: '30 1 * * *' 5 | 6 | jobs: 7 | stale: 8 | runs-on: ubuntu-latest 9 | steps: 10 | - uses: actions/stale@v9 11 | with: 12 | stale-issue-message: 'This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 5 days.' 13 | close-issue-message: 'This issue was closed because it has been stalled for 10 days with no activity.' 14 | days-before-stale: 60 15 | days-before-close: 10 16 | days-before-pr-close: -1 17 | -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | share/python-wheels/ 24 | *.egg-info/ 25 | .installed.cfg 26 | *.egg 27 | MANIFEST 28 | 29 | # PyInstaller 30 | # Usually these files are written by a python script from a template 31 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 32 | *.manifest 33 | *.spec 34 | 35 | # Installer logs 36 | pip-log.txt 37 | pip-delete-this-directory.txt 38 | 39 | # Unit test / coverage reports 40 | htmlcov/ 41 | .tox/ 42 | .nox/ 43 | .coverage 44 | .coverage.* 45 | .cache 46 | nosetests.xml 47 | coverage.xml 48 | *.cover 49 | *.py,cover 50 | .hypothesis/ 51 | .pytest_cache/ 52 | cover/ 53 | 54 | # Translations 55 | *.mo 56 | *.pot 57 | 58 | # Django stuff: 59 | *.log 60 | local_settings.py 61 | db.sqlite3 62 | db.sqlite3-journal 63 | 64 | # Flask stuff: 65 | instance/ 66 | .webassets-cache 67 | 68 | # Scrapy stuff: 69 | .scrapy 70 | 71 | # Sphinx documentation 72 | docs/_build/ 73 | 74 | # PyBuilder 75 | .pybuilder/ 76 | target/ 77 | 78 | # Jupyter Notebook 79 | .ipynb_checkpoints 80 | 81 | # IPython 82 | profile_default/ 83 | ipython_config.py 84 | 85 | # pyenv 86 | # For a library or package, you might want to ignore these files since the code is 87 | # intended to run in multiple environments; otherwise, check them in: 88 | # .python-version 89 | 90 | # pipenv 91 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 92 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 93 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 94 | # install all needed dependencies. 95 | #Pipfile.lock 96 | 97 | # poetry 98 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. 99 | # This is especially recommended for binary packages to ensure reproducibility, and is more 100 | # commonly ignored for libraries. 101 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control 102 | #poetry.lock 103 | 104 | # pdm 105 | # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control. 106 | #pdm.lock 107 | # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it 108 | # in version control. 109 | # https://pdm.fming.dev/#use-with-ide 110 | .pdm.toml 111 | 112 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm 113 | __pypackages__/ 114 | 115 | # Celery stuff 116 | celerybeat-schedule 117 | celerybeat.pid 118 | 119 | # SageMath parsed files 120 | *.sage.py 121 | 122 | # Environments 123 | .env 124 | .venv 125 | env/ 126 | venv/ 127 | ENV/ 128 | env.bak/ 129 | venv.bak/ 130 | 131 | # Spyder project settings 132 | .spyderproject 133 | .spyproject 134 | 135 | # Rope project settings 136 | .ropeproject 137 | 138 | # mkdocs documentation 139 | /site 140 | 141 | # mypy 142 | .mypy_cache/ 143 | .dmypy.json 144 | dmypy.json 145 | 146 | # Pyre type checker 147 | .pyre/ 148 | 149 | # pytype static type analyzer 150 | .pytype/ 151 | 152 | # Cython debug symbols 153 | cython_debug/ 154 | 155 | # PyCharm 156 | # JetBrains specific template is maintained in a separate JetBrains.gitignore that can 157 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore 158 | # and can be added to the global gitignore or merged into this file. For a more nuclear 159 | # option (not recommended) you can uncomment the following to ignore the entire idea folder. 160 | #.idea/ 161 | 162 | *.grib 163 | *.onnx 164 | *.ckpt 165 | *.swp 166 | *.npy 167 | *.download 168 | ? 169 | ?.* 170 | foo 171 | bar 172 | *.grib 173 | *.nc 174 | *.npz 175 | *.json 176 | *.req 177 | dev/ 178 | *.out 179 | _version.py 180 | *.tar 181 | -------------------------------------------------------------------------------- /.pre-commit-config.yaml: -------------------------------------------------------------------------------- 1 | repos: 2 | 3 | # Empty notebookds 4 | - repo: local 5 | hooks: 6 | - id: clear-notebooks-output 7 | name: clear-notebooks-output 8 | files: tools/.*\.ipynb$ 9 | stages: [commit] 10 | language: python 11 | entry: jupyter nbconvert --ClearOutputPreprocessor.enabled=True --inplace 12 | additional_dependencies: [jupyter] 13 | 14 | 15 | - repo: https://github.com/pre-commit/pre-commit-hooks 16 | rev: v4.4.0 17 | hooks: 18 | - id: check-yaml # Check YAML files for syntax errors only 19 | args: [--unsafe, --allow-multiple-documents] 20 | - id: debug-statements # Check for debugger imports and py37+ breakpoint() 21 | - id: end-of-file-fixer # Ensure files end in a newline 22 | - id: trailing-whitespace # Trailing whitespace checker 23 | - id: no-commit-to-branch # Prevent committing to main / master 24 | - id: check-added-large-files # Check for large files added to git 25 | - id: check-merge-conflict # Check for files that contain merge conflict 26 | 27 | - repo: https://github.com/psf/black-pre-commit-mirror 28 | rev: 24.1.1 29 | hooks: 30 | - id: black 31 | args: [--line-length=120] 32 | 33 | - repo: https://github.com/pycqa/isort 34 | rev: 5.13.2 35 | hooks: 36 | - id: isort 37 | args: 38 | - -l 120 39 | - --force-single-line-imports 40 | - --profile black 41 | 42 | 43 | - repo: https://github.com/astral-sh/ruff-pre-commit 44 | rev: v0.3.0 45 | hooks: 46 | - id: ruff 47 | exclude: '(dev/.*|.*_)\.py$' 48 | args: 49 | - --line-length=120 50 | - --fix 51 | - --exit-non-zero-on-fix 52 | - --preview 53 | 54 | - repo: https://github.com/sphinx-contrib/sphinx-lint 55 | rev: v0.9.1 56 | hooks: 57 | - id: sphinx-lint 58 | 59 | # For now, we use it. But it does not support a lot of sphinx features 60 | - repo: https://github.com/dzhu/rstfmt 61 | rev: v0.0.14 62 | hooks: 63 | - id: rstfmt 64 | 65 | - repo: https://github.com/b8raoult/pre-commit-docconvert 66 | rev: "0.1.4" 67 | hooks: 68 | - id: docconvert 69 | args: ["numpy"] 70 | -------------------------------------------------------------------------------- /CITATION.cff: -------------------------------------------------------------------------------- 1 | # This CITATION.cff file was generated with cffinit. 2 | # Visit https://bit.ly/cffinit to generate yours today! 3 | 4 | cff-version: 1.2.0 5 | title: ai-models 6 | message: >- 7 | If you use this software, please cite it using the 8 | metadata from this file. 9 | type: software 10 | authors: 11 | - given-names: Baudouin 12 | family-names: Raoult 13 | affiliation: ECMWF 14 | - given-names: Florian 15 | family-names: Pinault 16 | - given-names: Gert 17 | family-names: Mertes 18 | affiliation: ECMWF 19 | - given-names: Jesper Sören 20 | family-names: Dramsch 21 | affiliation: ECMWF 22 | orcid: 'https://orcid.org/0000-0001-8273-905X' 23 | - given-names: Harrison 24 | family-names: Cook 25 | affiliation: ECMWF 26 | orcid: 'https://orcid.org/0009-0009-3207-4876' 27 | - given-names: Matthew 28 | family-names: Chantry 29 | affiliation: ECMWF 30 | repository-code: 'https://github.com/ecmwf-lab/ai-models' 31 | abstract: >- 32 | ai-models is used to run AI-based weather forecasting 33 | models. These models need to be installed independently. 34 | keywords: 35 | - ai 36 | - weather forecasting 37 | license: Apache-2.0 38 | -------------------------------------------------------------------------------- /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 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # ai-models 2 | 3 |

4 | 5 | Static Badge 6 | 7 | 8 | 9 | Code Coverage 10 | 11 | 12 | 13 | License: Apache 2.0 14 | 15 | 16 | 17 | Latest Release 18 | 19 |

20 | 21 | **DISCLAIMER** 22 | 23 | > \[!IMPORTANT\] 24 | > This project is **BETA** and will be **Experimental** for the foreseeable future. 25 | > Interfaces and functionality are likely to change, and the project itself may be scrapped. 26 | > **DO NOT** use this software in any project/software that is operational. 27 | 28 | 29 | 30 | 31 | The `ai-models` command is used to run AI-based weather forecasting models. These models need to be installed independently. 32 | 33 | ## Usage 34 | 35 | Although the source code `ai-models` and its plugins are available under open sources licences, some model weights may be available under a different licence. For example some models make their weights available under the CC-BY-NC-SA 4.0 license, which does not allow commercial use. For more informations, please check the license associated with each model on their main home page, that we link from each of the corresponding plugins. 36 | 37 | ## Prerequisites 38 | 39 | Before using the `ai-models` command, ensure you have the following prerequisites: 40 | 41 | - Python 3.10 (it may work with different versions, but it has been tested with 3.10 on Linux/MacOS). 42 | - An ECMWF and/or CDS account for accessing input data (see below for more details). 43 | - A computed with a GPU for optimal performance (strongly recommended). 44 | 45 | ## Installation 46 | 47 | To install the `ai-models` command, run the following command: 48 | 49 | ```bash 50 | pip install ai-models 51 | ``` 52 | 53 | ## Available Models 54 | 55 | Currently, four models can be installed: 56 | 57 | ```bash 58 | pip install ai-models-panguweather 59 | pip install ai-models-fourcastnet 60 | pip install ai-models-graphcast # Install details at https://github.com/ecmwf-lab/ai-models-graphcast 61 | pip install ai-models-fourcastnetv2 62 | ``` 63 | 64 | See [ai-models-panguweather](https://github.com/ecmwf-lab/ai-models-panguweather), [ai-models-fourcastnet](https://github.com/ecmwf-lab/ai-models-fourcastnet), 65 | [ai-models-fourcastnetv2](https://github.com/ecmwf-lab/ai-models-fourcastnetv2) and [ai-models-graphcast](https://github.com/ecmwf-lab/ai-models-graphcast) for more details about these models. 66 | 67 | ## Running the models 68 | 69 | To run model, make sure it has been installed, then simply run: 70 | 71 | ```bash 72 | ai-models 73 | ``` 74 | 75 | Replace `` with the name of the specific AI model you want to run. 76 | 77 | By default, the model will be run for a 10-day lead time (240 hours), using yesterday's 12Z analysis from ECMWF's MARS archive. 78 | 79 | To produce a 15 days forecast, use the `--lead-time HOURS` option: 80 | 81 | ```bash 82 | ai-models --lead-time 360 83 | ``` 84 | 85 | You can change the other defaults using the available command line options, as described below. 86 | 87 | ## Performances Considerations 88 | 89 | The AI models can run on a CPU; however, they perform significantly better on a GPU. A 10-day forecast can take several hours on a CPU but only around one minute on a modern GPU. 90 | 91 | :warning: **We strongly recommend running these models on a computer equipped with a GPU for optimal performance.** 92 | 93 | It you see the following message when running a model, it means that the ONNX runtime was not able to find a the CUDA libraries on your system: 94 | > [W:onnxruntime:Default, onnxruntime_pybind_state.cc:541 CreateExecutionProviderInstance] Failed to create CUDAExecutionProvider. Please reference to ensure all dependencies are met. 95 | 96 | To fix this issue, we suggest that you install `ai-models` in a [conda](https://docs.conda.io/en/latest/) environment and install the CUDA libraries in that environment. For example: 97 | 98 | ```bash 99 | conda create -n ai-models python=3.10 100 | conda activate ai-models 101 | conda install cudatoolkit 102 | pip install ai-models 103 | ... 104 | ``` 105 | 106 | ## Assets 107 | 108 | The AI models rely on weights and other assets created during training. The first time you run a model, you will need to download the trained weights and any additional required assets. 109 | 110 | To download the assets before running a model, use the following command: 111 | 112 | ```bash 113 | ai-models --download-assets 114 | ``` 115 | 116 | The assets will be downloaded if needed and stored in the current directory. You can provide a different directory to store the assets: 117 | 118 | ```bash 119 | ai-models --download-assets --assets 120 | ``` 121 | 122 | Then, later on, simply use: 123 | 124 | ```bash 125 | ai-models --assets 126 | ``` 127 | 128 | or 129 | 130 | ```bash 131 | export AI_MODELS_ASSETS= 132 | ai-models 133 | ``` 134 | 135 | For better organisation of the assets directory, you can use the `--assets-sub-directory` option. This option will store the assets of each model in its own subdirectory within the specified assets directory. 136 | 137 | ## Input data 138 | 139 | The models require input data (initial conditions) to run. You can provide the input data using different sources, as described below: 140 | 141 | ### From MARS 142 | 143 | By default, `ai-models` use yesterday's 12Z analysis from ECMWF, fetched from the Centre's MARS archive using the [ECMWF WebAPI](https://www.ecmwf.int/en/computing/software/ecmwf-web-api). You will need an ECMWF account to access that service. 144 | 145 | To change the date or time, use the `--date` and `--time` options, respectively: 146 | 147 | ```bash 148 | ai-models --date YYYYMMDD --time HHMM 149 | ``` 150 | 151 | ### From the CDS 152 | 153 | You can start the models using ERA5 (ECMWF Reanalysis version 5) data for the [Copernicus Climate Data Store (CDS)](https://cds.climate.copernicus.eu/). You will need to create an account on the CDS. The data will be downloaded using the [CDS API](https://cds.climate.copernicus.eu/api-how-to). 154 | 155 | To access the CDS, simply add `--input cds` on the command line. Please note that ERA5 data is added to the CDS with a delay, so you will also have to provide a date with `--date YYYYMMDD`. 156 | 157 | ```bash 158 | ai-models --input cds --date 20230110 --time 0000 159 | ``` 160 | 161 | ### From a GRIB file 162 | 163 | If you have input data in the GRIB format, you can provide the file using the `--file` option: 164 | 165 | ```bash 166 | ai-models --file 167 | ``` 168 | 169 | The GRIB file can contain more fields than the ones required by the model. The `ai-models` command will automatically select the necessary fields from the file. 170 | 171 | To find out the list of fields needed by a specific model as initial conditions, use the following command: 172 | 173 | ```bash 174 | ai-models --fields 175 | ``` 176 | 177 | ## Output 178 | 179 | By default, the model output will be written in GRIB format in a file called `.grib`. You can change the file name with the option `--path `. If the path you specify contains placeholders between `{` and `}`, multiple files will be created based on the [eccodes](https://confluence.ecmwf.int/display/ECC) keys. For example: 180 | 181 | ```bash 182 | ai-models --path 'out-{step}.grib' 183 | ``` 184 | 185 | This command will create a file for each forecasted time step. 186 | 187 | If you want to disable writing the output to a file, use the `--output none` option. 188 | 189 | ## Command line options 190 | 191 | It has the following options: 192 | 193 | - `--help`: Displays this help message. 194 | - `--models`: Lists all installed models. 195 | - `--debug`: Turns on debug mode. This will print additional information to the console. 196 | 197 | ### Input 198 | 199 | - `--input INPUT`: The input source for the model. This can be a `mars`, `cds` or `file`. 200 | - `--file FILE`: The specific file to use as input. This option will set `--source` to `file`. 201 | 202 | - `--date DATE`: The analysis date for the model. This defaults to yesterday. 203 | - `--time TIME`: The analysis time for the model. This defaults to 1200. 204 | 205 | ### Output 206 | 207 | - `--output OUTPUT`: The output destination for the model. Values are `file` or `none`. 208 | - `--path PATH`: The path to write the output of the model. 209 | 210 | ### Run 211 | 212 | - `--lead-time HOURS`: The number of hours to forecast. The default is 240 (10 days). 213 | 214 | ### Assets management 215 | 216 | - `--assets ASSETS`: Specifies the path to the directory containing the model assets. The default is the current directory, but you can override it by setting the `$AI_MODELS_ASSETS` environment variable. 217 | - `--assets-sub-directory`: Enables organising assets in `/` subdirectories. 218 | - `--download-assets`: Downloads the assets if they do not exist. 219 | 220 | ### Misc. options 221 | 222 | - `--fields`: Print the list of fields needed by a model as initial conditions. 223 | - `--expver EXPVER`: The experiment version of the model output. 224 | - `--class CLASS`: The 'class' metadata of the model output. 225 | - `--metadata KEY=VALUE`: Additional metadata metadata in the model output 226 | 227 | ## License 228 | 229 | ``` 230 | Copyright 2022, European Centre for Medium Range Weather Forecasts. 231 | 232 | Licensed under the Apache License, Version 2.0 (the "License"); 233 | you may not use this file except in compliance with the License. 234 | You may obtain a copy of the License at 235 | 236 | http://www.apache.org/licenses/LICENSE-2.0 237 | 238 | Unless required by applicable law or agreed to in writing, software 239 | distributed under the License is distributed on an "AS IS" BASIS, 240 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 241 | See the License for the specific language governing permissions and 242 | limitations under the License. 243 | 244 | In applying this licence, ECMWF does not waive the privileges and immunities 245 | granted to it by virtue of its status as an intergovernmental organisation 246 | nor does it submit to any jurisdiction. 247 | ``` 248 | -------------------------------------------------------------------------------- /pyproject.toml: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # (C) Copyright 2024 ECMWF. 3 | # 4 | # This software is licensed under the terms of the Apache Licence Version 2.0 5 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 6 | # In applying this licence, ECMWF does not waive the privileges and immunities 7 | # granted to it by virtue of its status as an intergovernmental organisation 8 | # nor does it submit to any jurisdiction. 9 | 10 | # https://packaging.python.org/en/latest/guides/writing-pyproject-toml/ 11 | 12 | [build-system] 13 | requires = ["setuptools>=60", "setuptools-scm>=8.0"] 14 | 15 | [project] 16 | description = "A package to run AI weather models." 17 | name = "ai-models" 18 | 19 | dynamic = ["version"] 20 | license = { file = "LICENSE" } 21 | requires-python = ">=3.9" 22 | 23 | authors = [ 24 | { name = "European Centre for Medium-Range Weather Forecasts (ECMWF)", email = "software.support@ecmwf.int" }, 25 | ] 26 | 27 | keywords = ["tools", "ai"] 28 | 29 | classifiers = [ 30 | "Development Status :: 4 - Beta", 31 | "Intended Audience :: Developers", 32 | "License :: OSI Approved :: Apache Software License", 33 | "Programming Language :: Python :: 3", 34 | "Programming Language :: Python :: 3.9", 35 | "Programming Language :: Python :: 3.10", 36 | "Programming Language :: Python :: 3.11", 37 | "Programming Language :: Python :: Implementation :: CPython", 38 | "Programming Language :: Python :: Implementation :: PyPy", 39 | "Operating System :: OS Independent", 40 | ] 41 | 42 | dependencies = [ 43 | "cdsapi", 44 | "earthkit-data>=0.11.3", 45 | "earthkit-meteo", 46 | "earthkit-regrid", 47 | "eccodes>=2.37", 48 | "ecmwf-api-client", 49 | "ecmwf-opendata", 50 | "entrypoints", 51 | "gputil", 52 | "multiurl", 53 | "numpy<2", 54 | "pyyaml", 55 | "requests", 56 | "tqdm", 57 | ] 58 | 59 | 60 | [project.urls] 61 | Homepage = "https://github.com/ecmwf-lab/ai-models/" 62 | Repository = "https://github.com/ecmwf-lab/ai-models/" 63 | Issues = "https://github.com/ecmwf-lab/ai-models/issues" 64 | 65 | [project.scripts] 66 | ai-models = "ai_models.__main__:main" 67 | 68 | [tool.setuptools_scm] 69 | version_file = "src/ai_models/_version.py" 70 | 71 | [project.entry-points."ai_models.input"] 72 | file = "ai_models.inputs.file:FileInput" 73 | mars = "ai_models.inputs.mars:MarsInput" 74 | cds = "ai_models.inputs.cds:CdsInput" 75 | ecmwf-open-data = "ai_models.inputs.opendata:OpenDataInput" 76 | opendata = "ai_models.inputs.opendata:OpenDataInput" 77 | 78 | [project.entry-points."ai_models.output"] 79 | file = "ai_models.outputs:FileOutput" 80 | none = "ai_models.outputs:NoneOutput" 81 | -------------------------------------------------------------------------------- /src/ai_models/__init__.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | from ._version import __version__ 9 | -------------------------------------------------------------------------------- /src/ai_models/__main__.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import argparse 9 | import logging 10 | import os 11 | import shlex 12 | import sys 13 | 14 | import earthkit.data as ekd 15 | 16 | from .inputs import available_inputs 17 | from .model import Timer 18 | from .model import available_models 19 | from .model import load_model 20 | from .outputs import available_outputs 21 | 22 | ekd.settings.set("cache-policy", "user") 23 | 24 | LOG = logging.getLogger(__name__) 25 | 26 | 27 | def _main(argv): 28 | parser = argparse.ArgumentParser() 29 | 30 | # See https://github.com/pytorch/pytorch/issues/77764 31 | os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1" 32 | 33 | parser.add_argument( 34 | "--models", 35 | action="store_true", 36 | help="List models and exit", 37 | ) 38 | 39 | parser.add_argument( 40 | "--debug", 41 | action="store_true", 42 | help="Turn on debug", 43 | ) 44 | 45 | parser.add_argument( 46 | "-v", 47 | "--verbose", 48 | action="count", 49 | default=0, 50 | help="Increase verbosity", 51 | ) 52 | 53 | parser.add_argument( 54 | "--retrieve-requests", 55 | help=("Print mars requests to stdout." "Use --requests-extra to extend or overide the requests. "), 56 | action="store_true", 57 | ) 58 | 59 | parser.add_argument( 60 | "--archive-requests", 61 | help=("Save mars archive requests to FILE." "Use --requests-extra to extend or overide the requests. "), 62 | metavar="FILE", 63 | ) 64 | 65 | parser.add_argument( 66 | "--requests-extra", 67 | help=("Extends the retrieve or archive requests with a list of key1=value1,key2=value."), 68 | ) 69 | 70 | parser.add_argument( 71 | "--json", 72 | action="store_true", 73 | help=("Dump the requests in JSON format."), 74 | ) 75 | 76 | parser.add_argument( 77 | "--retrieve-fields-type", 78 | help="Type of field to retrieve. To use with --retrieve-requests.", 79 | choices=["constants", "prognostics", "all"], 80 | default="all", 81 | ) 82 | 83 | parser.add_argument( 84 | "--retrieve-only-one-date", 85 | help="Only retrieve the last date/time. To use with --retrieve-requests.", 86 | action="store_true", 87 | ) 88 | 89 | parser.add_argument( 90 | "--dump-provenance", 91 | metavar="FILE", 92 | help=("Dump information for tracking provenance."), 93 | ) 94 | 95 | parser.add_argument( 96 | "--input", 97 | default="mars", 98 | help="Source to use", 99 | choices=sorted(available_inputs()), 100 | ) 101 | 102 | parser.add_argument( 103 | "--file", 104 | help="Source to use if source=file", 105 | ) 106 | 107 | parser.add_argument( 108 | "--output", 109 | default="file", 110 | help="Where to output the results", 111 | choices=sorted(available_outputs()), 112 | ) 113 | 114 | parser.add_argument( 115 | "--date", 116 | default="-1", 117 | help="For which analysis date to start the inference (default: yesterday)", 118 | ) 119 | 120 | parser.add_argument( 121 | "--time", 122 | type=int, 123 | default=12, 124 | help="For which analysis time to start the inference (default: 12)", 125 | ) 126 | 127 | parser.add_argument( 128 | "--assets", 129 | default=os.environ.get("AI_MODELS_ASSETS", "."), 130 | help="Path to directory containing the weights and other assets", 131 | ) 132 | 133 | parser.add_argument( 134 | "--assets-sub-directory", 135 | help="Load assets from a subdirectory of --assets based on the name of the model.", 136 | action=argparse.BooleanOptionalAction, 137 | ) 138 | 139 | parser.parse_args(["--no-assets-sub-directory"]) 140 | 141 | parser.add_argument( 142 | "--assets-list", 143 | help="List the assets used by the model", 144 | action="store_true", 145 | ) 146 | 147 | parser.add_argument( 148 | "--download-assets", 149 | help="Download assets if they do not exists.", 150 | action="store_true", 151 | ) 152 | 153 | parser.add_argument( 154 | "--path", 155 | help="Path where to write the output of the model", 156 | ) 157 | 158 | parser.add_argument( 159 | "--fields", 160 | help="Show the fields needed as input for the model", 161 | action="store_true", 162 | ) 163 | 164 | parser.add_argument( 165 | "--expver", 166 | help="Set the experiment version of the model output. Has higher priority than --metadata.", 167 | ) 168 | 169 | parser.add_argument( 170 | "--class", 171 | help="Set the 'class' metadata of the model output. Has higher priority than --metadata.", 172 | metavar="CLASS", 173 | dest="class_", 174 | ) 175 | 176 | parser.add_argument( 177 | "--metadata", 178 | help="Set additional metadata metadata in the model output", 179 | metavar="KEY=VALUE", 180 | action="append", 181 | ) 182 | 183 | parser.add_argument( 184 | "--num-threads", 185 | type=int, 186 | default=1, 187 | help="Number of threads. Only relevant for some models.", 188 | ) 189 | 190 | parser.add_argument( 191 | "--lead-time", 192 | type=int, 193 | default=240, 194 | help="Length of forecast in hours.", 195 | ) 196 | 197 | parser.add_argument( 198 | "--hindcast-reference-year", 199 | help="For encoding hincast-like outputs", 200 | ) 201 | 202 | parser.add_argument( 203 | "--hindcast-reference-date", 204 | help="For encoding hincast-like outputs", 205 | ) 206 | 207 | parser.add_argument( 208 | "--staging-dates", 209 | help="For encoding hincast-like outputs", 210 | ) 211 | 212 | parser.add_argument( 213 | "--only-gpu", 214 | help="Fail if GPU is not available", 215 | action="store_true", 216 | ) 217 | 218 | parser.add_argument( 219 | "--deterministic", 220 | help="Fail if GPU is not available", 221 | action="store_true", 222 | ) 223 | 224 | # TODO: deprecate that option 225 | parser.add_argument( 226 | "--model-version", 227 | default="latest", 228 | help="Model version", 229 | ) 230 | 231 | parser.add_argument( 232 | "--version", 233 | action="store_true", 234 | help="Print ai-models version and exit", 235 | ) 236 | 237 | if all(arg not in ("--models", "--version") for arg in argv): 238 | parser.add_argument( 239 | "model", 240 | metavar="MODEL", 241 | choices=available_models() if "--remote" not in argv else None, 242 | help="The model to run", 243 | ) 244 | 245 | parser.add_argument( 246 | "--remote", 247 | help="Enable remote execution, read url and token from ~/.config/ai-models/api.yaml", 248 | action="store_true", 249 | dest="remote_execution", 250 | default=(os.environ.get("AI_MODELS_REMOTE", "0") == "1"), 251 | ) 252 | 253 | args, unknownargs = parser.parse_known_args(argv) 254 | 255 | if args.version: 256 | from ai_models import __version__ 257 | 258 | print(__version__) 259 | sys.exit(0) 260 | 261 | del args.version 262 | 263 | if args.models: 264 | if args.remote_execution: 265 | from .remote import RemoteAPI 266 | 267 | api = RemoteAPI() 268 | models = api.models() 269 | if len(models) == 0: 270 | print(f"No remote models available on {api.url}") 271 | sys.exit(0) 272 | print(f"Models available on remote server {api.url}") 273 | else: 274 | models = available_models() 275 | 276 | for p in sorted(models): 277 | print(p) 278 | sys.exit(0) 279 | 280 | if args.assets_sub_directory: 281 | args.assets = os.path.join(args.assets, args.model) 282 | 283 | if args.path is None: 284 | args.path = f"{args.model}.grib" 285 | 286 | if args.file is not None: 287 | args.input = "file" 288 | 289 | if not args.fields and not args.retrieve_requests: 290 | logging.basicConfig( 291 | level="DEBUG" if args.debug else "INFO", 292 | format="%(asctime)s %(levelname)s %(message)s", 293 | ) 294 | 295 | if args.metadata is None: 296 | args.metadata = [] 297 | 298 | args.metadata = dict(kv.split("=") for kv in args.metadata) 299 | 300 | if args.expver is not None: 301 | args.metadata["expver"] = args.expver 302 | 303 | if args.class_ is not None: 304 | args.metadata["class"] = args.class_ 305 | 306 | if args.requests_extra: 307 | if not args.retrieve_requests and not args.archive_requests: 308 | parser.error("You need to specify --retrieve-requests or --archive-requests") 309 | 310 | run(vars(args), unknownargs) 311 | 312 | 313 | def run(cfg: dict, model_args: list): 314 | if cfg["remote_execution"]: 315 | from .remote import RemoteModel 316 | 317 | model = RemoteModel(**cfg, model_args=model_args) 318 | else: 319 | model = load_model(cfg["model"], **cfg, model_args=model_args) 320 | 321 | if cfg["fields"]: 322 | model.print_fields() 323 | sys.exit(0) 324 | 325 | # This logic is a bit convoluted, but it is for backwards compatibility. 326 | if cfg["retrieve_requests"] or (cfg["requests_extra"] and not cfg["archive_requests"]): 327 | model.print_requests() 328 | sys.exit(0) 329 | 330 | if cfg["assets_list"]: 331 | model.print_assets_list() 332 | sys.exit(0) 333 | 334 | try: 335 | model.run() 336 | except FileNotFoundError as e: 337 | LOG.exception(e) 338 | LOG.error( 339 | "It is possible that some files required by %s are missing.", 340 | cfg["model"], 341 | ) 342 | LOG.error("Rerun the command as:") 343 | LOG.error( 344 | " %s", 345 | shlex.join([sys.argv[0], "--download-assets"] + sys.argv[1:]), 346 | ) 347 | sys.exit(1) 348 | 349 | model.finalise() 350 | 351 | if cfg["dump_provenance"]: 352 | with Timer("Collect provenance information"): 353 | with open(cfg["dump_provenance"], "w") as f: 354 | prov = model.provenance() 355 | import json # import here so it is not listed in provenance 356 | 357 | json.dump(prov, f, indent=4) 358 | 359 | 360 | def main(): 361 | with Timer("Total time"): 362 | _main(sys.argv[1:]) 363 | 364 | 365 | if __name__ == "__main__": 366 | main() 367 | -------------------------------------------------------------------------------- /src/ai_models/checkpoint.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | import os 10 | import pickle 11 | import zipfile 12 | from typing import Any 13 | 14 | LOG = logging.getLogger(__name__) 15 | 16 | 17 | class FakeStorage: 18 | def __init__(self): 19 | import torch 20 | 21 | self.dtype = torch.float32 22 | self._untyped_storage = torch.UntypedStorage(0) 23 | 24 | 25 | class UnpicklerWrapper(pickle.Unpickler): 26 | def __init__(self, file, **kwargs): 27 | super().__init__(file, **kwargs) 28 | 29 | def persistent_load(self, pid: Any) -> Any: 30 | return FakeStorage() 31 | 32 | 33 | def tidy(x): 34 | if isinstance(x, dict): 35 | return {k: tidy(v) for k, v in x.items()} 36 | 37 | if isinstance(x, list): 38 | return [tidy(v) for v in x] 39 | 40 | if isinstance(x, tuple): 41 | return tuple([tidy(v) for v in x]) 42 | 43 | if x is None: 44 | return None 45 | 46 | if isinstance(x, (int, float, str, bool)): 47 | return x 48 | 49 | return x 50 | 51 | 52 | def peek(path): 53 | with zipfile.ZipFile(path, "r") as f: 54 | data_pkl = None 55 | for b in f.namelist(): 56 | if os.path.basename(b) == "data.pkl": 57 | if data_pkl is not None: 58 | raise Exception(f"Found two data.pkl files in {path}: {data_pkl} and {b}") 59 | data_pkl = b 60 | 61 | LOG.info(f"Found data.pkl at {data_pkl}") 62 | 63 | with zipfile.ZipFile(path, "r") as f: 64 | unpickler = UnpicklerWrapper(f.open(data_pkl, "r")) 65 | x = tidy(unpickler.load()) 66 | return tidy(x) 67 | -------------------------------------------------------------------------------- /src/ai_models/inputs/__init__.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | from functools import cached_property 10 | 11 | import earthkit.data as ekd 12 | import earthkit.regrid as ekr 13 | import entrypoints 14 | from earthkit.data.indexing.fieldlist import FieldArray 15 | 16 | LOG = logging.getLogger(__name__) 17 | 18 | 19 | def get_input(name, *args, **kwargs): 20 | return available_inputs()[name].load()(*args, **kwargs) 21 | 22 | 23 | def available_inputs(): 24 | result = {} 25 | for e in entrypoints.get_group_all("ai_models.input"): 26 | result[e.name] = e 27 | return result 28 | -------------------------------------------------------------------------------- /src/ai_models/inputs/base.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | from functools import cached_property 10 | 11 | import earthkit.data as ekd 12 | 13 | LOG = logging.getLogger(__name__) 14 | 15 | 16 | class RequestBasedInput: 17 | def __init__(self, owner, **kwargs): 18 | self.owner = owner 19 | 20 | def _patch(self, **kargs): 21 | r = dict(**kargs) 22 | self.owner.patch_retrieve_request(r) 23 | return r 24 | 25 | @cached_property 26 | def fields_sfc(self): 27 | param = self.owner.param_sfc 28 | if not param: 29 | return ekd.from_source("empty") 30 | 31 | LOG.info(f"Loading surface fields from {self.WHERE}") 32 | 33 | return ekd.from_source( 34 | "multi", 35 | [ 36 | self.sfc_load_source( 37 | **self._patch( 38 | date=date, 39 | time=time, 40 | param=param, 41 | grid=self.owner.grid, 42 | area=self.owner.area, 43 | **self.owner.retrieve, 44 | ) 45 | ) 46 | for date, time in self.owner.datetimes() 47 | ], 48 | ) 49 | 50 | @cached_property 51 | def fields_pl(self): 52 | param, level = self.owner.param_level_pl 53 | if not (param and level): 54 | return ekd.from_source("empty") 55 | 56 | LOG.info(f"Loading pressure fields from {self.WHERE}") 57 | return ekd.from_source( 58 | "multi", 59 | [ 60 | self.pl_load_source( 61 | **self._patch( 62 | date=date, 63 | time=time, 64 | param=param, 65 | level=level, 66 | grid=self.owner.grid, 67 | area=self.owner.area, 68 | ) 69 | ) 70 | for date, time in self.owner.datetimes() 71 | ], 72 | ) 73 | 74 | @cached_property 75 | def fields_ml(self): 76 | param, level = self.owner.param_level_ml 77 | if not (param and level): 78 | return ekd.from_source("empty") 79 | 80 | LOG.info(f"Loading model fields from {self.WHERE}") 81 | return ekd.from_source( 82 | "multi", 83 | [ 84 | self.ml_load_source( 85 | **self._patch( 86 | date=date, 87 | time=time, 88 | param=param, 89 | level=level, 90 | grid=self.owner.grid, 91 | area=self.owner.area, 92 | ) 93 | ) 94 | for date, time in self.owner.datetimes() 95 | ], 96 | ) 97 | 98 | @cached_property 99 | def all_fields(self): 100 | return self.fields_sfc + self.fields_pl + self.fields_ml 101 | -------------------------------------------------------------------------------- /src/ai_models/inputs/cds.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | import earthkit.data as ekd 11 | 12 | from .base import RequestBasedInput 13 | 14 | LOG = logging.getLogger(__name__) 15 | 16 | 17 | class CdsInput(RequestBasedInput): 18 | WHERE = "CDS" 19 | 20 | def pl_load_source(self, **kwargs): 21 | kwargs["product_type"] = "reanalysis" 22 | return ekd.from_source("cds", "reanalysis-era5-pressure-levels", kwargs) 23 | 24 | def sfc_load_source(self, **kwargs): 25 | kwargs["product_type"] = "reanalysis" 26 | return ekd.from_source("cds", "reanalysis-era5-single-levels", kwargs) 27 | 28 | def ml_load_source(self, **kwargs): 29 | raise NotImplementedError("CDS does not support model levels") 30 | -------------------------------------------------------------------------------- /src/ai_models/inputs/compute.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2024 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | import earthkit.data as ekd 11 | import tqdm 12 | from earthkit.data.core.temporary import temp_file 13 | from earthkit.data.indexing.fieldlist import FieldArray 14 | 15 | LOG = logging.getLogger(__name__) 16 | 17 | G = 9.80665 # Same a pgen 18 | 19 | 20 | def make_z_from_gh(ds): 21 | 22 | tmp = temp_file() 23 | 24 | out = ekd.new_grib_output(tmp.path) 25 | other = [] 26 | 27 | for f in tqdm.tqdm(ds, delay=0.5, desc="GH to Z", leave=False): 28 | 29 | if f.metadata("param") == "gh": 30 | out.write(f.to_numpy() * G, template=f, param="z") 31 | else: 32 | other.append(f) 33 | 34 | out.close() 35 | 36 | result = FieldArray(other) + ekd.from_source("file", tmp.path) 37 | result._tmp = tmp 38 | 39 | return result 40 | -------------------------------------------------------------------------------- /src/ai_models/inputs/file.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | from functools import cached_property 10 | 11 | import earthkit.data as ekd 12 | import entrypoints 13 | 14 | LOG = logging.getLogger(__name__) 15 | 16 | 17 | class FileInput: 18 | def __init__(self, owner, file, **kwargs): 19 | self.file = file 20 | self.owner = owner 21 | 22 | @cached_property 23 | def fields_sfc(self): 24 | return self.all_fields.sel(levtype="sfc") 25 | 26 | @cached_property 27 | def fields_pl(self): 28 | return self.all_fields.sel(levtype="pl") 29 | 30 | @cached_property 31 | def fields_ml(self): 32 | return self.all_fields.sel(levtype="ml") 33 | 34 | @cached_property 35 | def all_fields(self): 36 | return ekd.from_source("file", self.file) 37 | 38 | 39 | def get_input(name, *args, **kwargs): 40 | return available_inputs()[name].load()(*args, **kwargs) 41 | 42 | 43 | def available_inputs(): 44 | result = {} 45 | for e in entrypoints.get_group_all("ai_models.input"): 46 | result[e.name] = e 47 | return result 48 | -------------------------------------------------------------------------------- /src/ai_models/inputs/interpolate.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2024 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | import earthkit.data as ekd 11 | import earthkit.regrid as ekr 12 | import tqdm 13 | from earthkit.data.core.temporary import temp_file 14 | 15 | LOG = logging.getLogger(__name__) 16 | 17 | 18 | class Interpolate: 19 | def __init__(self, grid, source, metadata): 20 | self.grid = list(grid) if isinstance(grid, tuple) else grid 21 | self.source = list(source) if isinstance(source, tuple) else source 22 | self.metadata = metadata 23 | 24 | def __call__(self, ds): 25 | tmp = temp_file() 26 | 27 | out = ekd.new_grib_output(tmp.path) 28 | 29 | result = [] 30 | for f in tqdm.tqdm(ds, delay=0.5, desc="Interpolating", leave=False): 31 | data = ekr.interpolate(f.to_numpy(), dict(grid=self.source), dict(grid=self.grid)) 32 | out.write(data, template=f, **self.metadata) 33 | 34 | out.close() 35 | 36 | result = ekd.from_source("file", tmp.path) 37 | result._tmp = tmp 38 | 39 | print("Interpolated data", tmp.path) 40 | 41 | return result 42 | -------------------------------------------------------------------------------- /src/ai_models/inputs/mars.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | import earthkit.data as ekd 11 | 12 | from .base import RequestBasedInput 13 | 14 | LOG = logging.getLogger(__name__) 15 | 16 | 17 | class MarsInput(RequestBasedInput): 18 | WHERE = "MARS" 19 | 20 | def __init__(self, owner, **kwargs): 21 | self.owner = owner 22 | 23 | def pl_load_source(self, **kwargs): 24 | kwargs["levtype"] = "pl" 25 | logging.debug("load source mars %s", kwargs) 26 | return ekd.from_source("mars", kwargs) 27 | 28 | def sfc_load_source(self, **kwargs): 29 | kwargs["levtype"] = "sfc" 30 | logging.debug("load source mars %s", kwargs) 31 | return ekd.from_source("mars", kwargs) 32 | 33 | def ml_load_source(self, **kwargs): 34 | kwargs["levtype"] = "ml" 35 | logging.debug("load source mars %s", kwargs) 36 | return ekd.from_source("mars", kwargs) 37 | -------------------------------------------------------------------------------- /src/ai_models/inputs/opendata.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2024 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import itertools 9 | import logging 10 | import os 11 | 12 | import earthkit.data as ekd 13 | from earthkit.data.core.temporary import temp_file 14 | from earthkit.data.indexing.fieldlist import FieldArray 15 | from multiurl import download 16 | 17 | from .base import RequestBasedInput 18 | from .compute import make_z_from_gh 19 | from .interpolate import Interpolate 20 | from .recenter import recenter 21 | from .transform import NewMetadataField 22 | 23 | LOG = logging.getLogger(__name__) 24 | 25 | CONSTANTS = ( 26 | "z", 27 | "sdor", 28 | "slor", 29 | ) 30 | 31 | CONSTANTS_URL = "https://get.ecmwf.int/repository/test-data/ai-models/opendata/constants-{resol}.grib2" 32 | 33 | RESOLS = { 34 | (0.25, 0.25): ("0p25", (0.25, 0.25), False, False, {}), 35 | (0.1, 0.1): ( 36 | "0p25", 37 | (0.25, 0.25), 38 | True, 39 | True, 40 | dict( 41 | longitudeOfLastGridPointInDegrees=359.9, 42 | iDirectionIncrementInDegrees=0.1, 43 | jDirectionIncrementInDegrees=0.1, 44 | Ni=3600, 45 | Nj=1801, 46 | ), 47 | ), 48 | # "N320": ("0p25", (0.25, 0.25), True, False, dict(gridType='reduced_gg')), 49 | # "O96": ("0p25", (0.25, 0.25), True, False, dict(gridType='reduced_gg', )), 50 | } 51 | 52 | 53 | def _identity(x): 54 | return x 55 | 56 | 57 | class OpenDataInput(RequestBasedInput): 58 | WHERE = "OPENDATA" 59 | 60 | def __init__(self, owner, **kwargs): 61 | self.owner = owner 62 | 63 | def _adjust(self, kwargs): 64 | 65 | kwargs.setdefault("step", 0) 66 | 67 | if "level" in kwargs: 68 | # OpenData uses levelist instead of level 69 | kwargs["levelist"] = kwargs.pop("level") 70 | 71 | if "area" in kwargs: 72 | kwargs.pop("area") 73 | 74 | grid = kwargs.pop("grid") 75 | if isinstance(grid, list): 76 | grid = tuple(grid) 77 | 78 | kwargs["resol"], source, interp, oversampling, metadata = RESOLS[grid] 79 | r = dict(**kwargs) 80 | r.update(self.owner.retrieve) 81 | 82 | if interp: 83 | 84 | logging.info("Interpolating input data from %s to %s.", source, grid) 85 | if oversampling: 86 | logging.warning("This will oversample the input data.") 87 | return Interpolate(grid, source, metadata) 88 | else: 89 | return _identity 90 | 91 | def pl_load_source(self, **kwargs): 92 | 93 | gh_to_z = _identity 94 | interpolate = self._adjust(kwargs) 95 | 96 | kwargs["levtype"] = "pl" 97 | request = kwargs.copy() 98 | 99 | param = [p.lower() for p in kwargs["param"]] 100 | assert isinstance(param, (list, tuple)) 101 | 102 | if "z" in param: 103 | logging.warning("Parameter 'z' on pressure levels is not available in ECMWF open data, using 'gh' instead") 104 | param = list(param) 105 | param.remove("z") 106 | if "gh" not in param: 107 | param.append("gh") 108 | kwargs["param"] = param 109 | gh_to_z = make_z_from_gh 110 | 111 | logging.info("ecmwf-open-data %s", kwargs) 112 | 113 | opendata = recenter(ekd.from_source("ecmwf-open-data", **kwargs)) 114 | opendata = gh_to_z(opendata) 115 | opendata = interpolate(opendata) 116 | 117 | return self.check_pl(opendata, request) 118 | 119 | def constants(self, constant_params, request, kwargs): 120 | if len(constant_params) == 1: 121 | logging.warning( 122 | f"Single level parameter '{constant_params[0]}' is" 123 | " not available in ECMWF open data, using constants.grib2 instead" 124 | ) 125 | else: 126 | logging.warning( 127 | f"Single level parameters {constant_params} are" 128 | " not available in ECMWF open data, using constants.grib2 instead" 129 | ) 130 | 131 | cachedir = os.path.expanduser("~/.cache/ai-models") 132 | constants_url = CONSTANTS_URL.format(resol=request["resol"]) 133 | basename = os.path.basename(constants_url) 134 | 135 | if not os.path.exists(cachedir): 136 | os.makedirs(cachedir) 137 | 138 | path = os.path.join(cachedir, basename) 139 | 140 | if not os.path.exists(path): 141 | logging.info("Downloading %s to %s", constants_url, path) 142 | download(constants_url, path + ".tmp") 143 | os.rename(path + ".tmp", path) 144 | 145 | ds = ekd.from_source("file", path) 146 | ds = ds.sel(param=constant_params) 147 | 148 | tmp = temp_file() 149 | 150 | out = ekd.new_grib_output(tmp.path) 151 | 152 | for f in ds: 153 | out.write( 154 | f.to_numpy(), 155 | template=f, 156 | date=kwargs["date"], 157 | time=kwargs["time"], 158 | step=kwargs.get("step", 0), 159 | ) 160 | 161 | out.close() 162 | 163 | result = ekd.from_source("file", tmp.path) 164 | result._tmp = tmp 165 | 166 | return result 167 | 168 | def sfc_load_source(self, **kwargs): 169 | interpolate = self._adjust(kwargs) 170 | 171 | kwargs["levtype"] = "sfc" 172 | request = kwargs.copy() 173 | 174 | param = [p.lower() for p in kwargs["param"]] 175 | assert isinstance(param, (list, tuple)) 176 | 177 | constant_params = [] 178 | param = list(param) 179 | for c in CONSTANTS: 180 | if c in param: 181 | param.remove(c) 182 | constant_params.append(c) 183 | 184 | if constant_params: 185 | constants = self.constants(constant_params, request, kwargs) 186 | else: 187 | constants = ekd.from_source("empty") 188 | 189 | kwargs["param"] = param 190 | 191 | opendata = recenter(ekd.from_source("ecmwf-open-data", **kwargs)) 192 | opendata = opendata + constants 193 | opendata = interpolate(opendata) 194 | 195 | # Fix grib2/eccodes bug 196 | 197 | opendata = FieldArray([NewMetadataField(f, levelist=None) for f in opendata]) 198 | 199 | return self.check_sfc(opendata, request) 200 | 201 | def ml_load_source(self, **kwargs): 202 | interpolate = self._adjust(kwargs) 203 | kwargs["levtype"] = "ml" 204 | request = kwargs.copy() 205 | 206 | opendata = recenter(ekd.from_source("ecmwf-open-data", **kwargs)) 207 | opendata = interpolate(opendata) 208 | 209 | return self.check_ml(opendata, request) 210 | 211 | def check_pl(self, ds, request): 212 | self._check(ds, "PL", request, "param", "levelist") 213 | return ds 214 | 215 | def check_sfc(self, ds, request): 216 | self._check(ds, "SFC", request, "param") 217 | return ds 218 | 219 | def check_ml(self, ds, request): 220 | self._check(ds, "ML", request, "param", "levelist") 221 | return ds 222 | 223 | def _check(self, ds, what, request, *keys): 224 | 225 | def _(p): 226 | if len(p) == 1: 227 | return p[0] 228 | 229 | expected = set() 230 | for p in itertools.product(*[request[key] for key in keys]): 231 | expected.add(p) 232 | 233 | found = set() 234 | for f in ds: 235 | found.add(tuple(f.metadata(key) for key in keys)) 236 | 237 | missing = expected - found 238 | if missing: 239 | missing = [_(p) for p in missing] 240 | if len(missing) == 1: 241 | raise ValueError(f"The following {what} parameter '{missing[0]}' is not available in ECMWF open data") 242 | raise ValueError(f"The following {what} parameters {missing} are not available in ECMWF open data") 243 | 244 | extra = found - expected 245 | if extra: 246 | extra = [_(p) for p in extra] 247 | if len(extra) == 1: 248 | raise ValueError(f"Unexpected {what} parameter '{extra[0]}' from ECMWF open data") 249 | raise ValueError(f"Unexpected {what} parameters {extra} from ECMWF open data") 250 | -------------------------------------------------------------------------------- /src/ai_models/inputs/recenter.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2024 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | import earthkit.data as ekd 11 | import numpy as np 12 | import tqdm 13 | from earthkit.data.core.temporary import temp_file 14 | 15 | LOG = logging.getLogger(__name__) 16 | 17 | CHECKED = set() 18 | 19 | 20 | def _init_recenter(ds, f): 21 | 22 | # For now, we only support the 0.25x0.25 grid from OPENDATA (centered on the greenwich meridian) 23 | 24 | latitudeOfFirstGridPointInDegrees = f.metadata("latitudeOfFirstGridPointInDegrees") 25 | longitudeOfFirstGridPointInDegrees = f.metadata("longitudeOfFirstGridPointInDegrees") 26 | latitudeOfLastGridPointInDegrees = f.metadata("latitudeOfLastGridPointInDegrees") 27 | longitudeOfLastGridPointInDegrees = f.metadata("longitudeOfLastGridPointInDegrees") 28 | iDirectionIncrementInDegrees = f.metadata("iDirectionIncrementInDegrees") 29 | jDirectionIncrementInDegrees = f.metadata("jDirectionIncrementInDegrees") 30 | scanningMode = f.metadata("scanningMode") 31 | Ni = f.metadata("Ni") 32 | Nj = f.metadata("Nj") 33 | 34 | assert scanningMode == 0 35 | assert latitudeOfFirstGridPointInDegrees == 90 36 | assert longitudeOfFirstGridPointInDegrees == 180 37 | assert latitudeOfLastGridPointInDegrees == -90 38 | assert longitudeOfLastGridPointInDegrees == 179.75 39 | assert iDirectionIncrementInDegrees == 0.25 40 | assert jDirectionIncrementInDegrees == 0.25 41 | 42 | assert Ni == 1440 43 | assert Nj == 721 44 | 45 | shape = (Nj, Ni) 46 | roll = -Ni // 2 47 | axis = 1 48 | 49 | key = ( 50 | latitudeOfFirstGridPointInDegrees, 51 | longitudeOfFirstGridPointInDegrees, 52 | latitudeOfLastGridPointInDegrees, 53 | longitudeOfLastGridPointInDegrees, 54 | iDirectionIncrementInDegrees, 55 | jDirectionIncrementInDegrees, 56 | Ni, 57 | Nj, 58 | ) 59 | 60 | ############################ 61 | 62 | if key not in CHECKED: 63 | lon = ekd.from_source("forcings", ds, param=["longitude"], date=f.metadata("date"))[0] 64 | assert np.all(np.roll(lon.to_numpy(), roll, axis=axis)[:, 0] == 0) 65 | CHECKED.add(key) 66 | 67 | return (shape, roll, axis, dict(longitudeOfFirstGridPointInDegrees=0, longitudeOfLastGridPointInDegrees=359.75)) 68 | 69 | 70 | def recenter(ds): 71 | 72 | tmp = temp_file() 73 | 74 | out = ekd.new_grib_output(tmp.path) 75 | 76 | for f in tqdm.tqdm(ds, delay=0.5, desc="Recentering", leave=False): 77 | 78 | shape, roll, axis, metadata = _init_recenter(ds, f) 79 | 80 | data = f.to_numpy() 81 | assert data.shape == shape, (data.shape, shape) 82 | 83 | data = np.roll(data, roll, axis=axis) 84 | 85 | out.write(data, template=f, **metadata) 86 | 87 | out.close() 88 | 89 | result = ekd.from_source("file", tmp.path) 90 | result._tmp = tmp 91 | 92 | return result 93 | -------------------------------------------------------------------------------- /src/ai_models/inputs/transform.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2024 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | 10 | LOG = logging.getLogger(__name__) 11 | 12 | 13 | class WrappedField: 14 | def __init__(self, field): 15 | self._field = field 16 | 17 | def __getattr__(self, name): 18 | return getattr(self._field, name) 19 | 20 | def __repr__(self) -> str: 21 | return repr(self._field) 22 | 23 | 24 | class NewDataField(WrappedField): 25 | def __init__(self, field, data): 26 | super().__init__(field) 27 | self._data = data 28 | self.shape = data.shape 29 | 30 | def to_numpy(self, flatten=False, dtype=None, index=None): 31 | data = self._data 32 | if dtype is not None: 33 | data = data.astype(dtype) 34 | if flatten: 35 | data = data.flatten() 36 | if index is not None: 37 | data = data[index] 38 | return data 39 | 40 | 41 | class NewMetadataField(WrappedField): 42 | def __init__(self, field, **kwargs): 43 | super().__init__(field) 44 | self._metadata = kwargs 45 | 46 | def metadata(self, *args, **kwargs): 47 | if len(args) == 1 and args[0] in self._metadata: 48 | return self._metadata[args[0]] 49 | return self._field.metadata(*args, **kwargs) 50 | -------------------------------------------------------------------------------- /src/ai_models/model.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import base64 9 | import datetime 10 | import json 11 | import logging 12 | import os 13 | import sys 14 | import time 15 | from collections import defaultdict 16 | from functools import cached_property 17 | 18 | import earthkit.data as ekd 19 | import entrypoints 20 | import numpy as np 21 | from earthkit.data.utils.humanize import seconds 22 | from multiurl import download 23 | 24 | from .checkpoint import peek 25 | from .inputs import get_input 26 | from .outputs import get_output 27 | from .stepper import Stepper 28 | 29 | LOG = logging.getLogger(__name__) 30 | 31 | 32 | class Timer: 33 | def __init__(self, title): 34 | self.title = title 35 | self.start = time.time() 36 | 37 | def __enter__(self): 38 | return self 39 | 40 | def __exit__(self, *args): 41 | elapsed = time.time() - self.start 42 | LOG.info("%s: %s.", self.title, seconds(elapsed)) 43 | 44 | 45 | class ArchiveCollector: 46 | UNIQUE = {"date", "hdate", "time", "referenceDate", "type", "stream", "expver"} 47 | 48 | def __init__(self) -> None: 49 | self.expect = 0 50 | self.request = defaultdict(set) 51 | 52 | def add(self, field): 53 | self.expect += 1 54 | for k, v in field.items(): 55 | self.request[k].add(str(v)) 56 | if k in self.UNIQUE: 57 | if len(self.request[k]) > 1: 58 | raise ValueError(f"Field {field} has different values for {k}: {self.request[k]}") 59 | 60 | 61 | class Model: 62 | lagged = False 63 | assets_extra_dir = None 64 | retrieve = {} # Extra parameters for retrieve 65 | version = 1 # To be overriden in subclasses 66 | grib_edition = 2 # Default GRIB edition 67 | grib_extra_metadata = {} # Extra metadata for grib files 68 | 69 | param_level_ml = ([], []) # param, level 70 | param_level_pl = ([], []) # param, level 71 | param_sfc = [] # param 72 | 73 | def __init__(self, input, output, download_assets, **kwargs): 74 | self.input = get_input(input, self, **kwargs) 75 | self.output = get_output(output, self, **kwargs) 76 | 77 | for k, v in kwargs.items(): 78 | setattr(self, k, v) 79 | 80 | # We need to call it to initialise the default args 81 | args = self.parse_model_args(self.model_args) 82 | if args: 83 | for k, v in vars(args).items(): 84 | setattr(self, k, v) 85 | 86 | if self.assets_sub_directory: 87 | if self.assets_extra_dir is not None: 88 | self.assets += self.assets_extra_dir 89 | 90 | LOG.debug("Asset directory is %s", self.assets) 91 | 92 | try: 93 | self.date = int(self.date) 94 | except ValueError: 95 | pass 96 | 97 | if download_assets: 98 | self.download_assets(**kwargs) 99 | 100 | self.archiving = defaultdict(ArchiveCollector) 101 | self.created = time.time() 102 | 103 | @cached_property 104 | def fields_pl(self): 105 | return self.input.fields_pl 106 | 107 | @cached_property 108 | def fields_ml(self): 109 | return self.input.fields_ml 110 | 111 | @cached_property 112 | def fields_sfc(self): 113 | return self.input.fields_sfc 114 | 115 | @cached_property 116 | def all_fields(self): 117 | return self.input.all_fields 118 | 119 | def write(self, *args, **kwargs): 120 | self.collect_archive_requests( 121 | self.output.write(*args, **kwargs, **self.grib_extra_metadata), 122 | ) 123 | 124 | def collect_archive_requests(self, written): 125 | if self.archive_requests: 126 | handle, path = written 127 | if self.hindcast_reference_year or self.hindcast_reference_date: 128 | # The clone is necessary because the handle 129 | # does not return always return recently set keys 130 | handle = handle.clone() 131 | 132 | self.archiving[path].add(handle.as_namespace("mars")) 133 | 134 | def finalise(self): 135 | self.output.flush() 136 | 137 | if self.archive_requests: 138 | with open(self.archive_requests, "w") as f: 139 | json_requests = [] 140 | 141 | for path, archive in self.archiving.items(): 142 | request = dict(expect=archive.expect) 143 | if path is not None: 144 | request["source"] = f'"{path}"' 145 | request.update(archive.request) 146 | request.update(self._requests_extra) 147 | 148 | if self.json: 149 | json_requests.append(request) 150 | else: 151 | self._print_request("archive", request, file=f) 152 | 153 | if json_requests: 154 | 155 | def json_default(obj): 156 | if isinstance(obj, set): 157 | if len(obj) > 1: 158 | return sorted(list(obj)) 159 | else: 160 | return obj.pop() 161 | raise TypeError 162 | 163 | print( 164 | json.dumps(json_requests, separators=(",", ":"), default=json_default, sort_keys=True), 165 | file=f, 166 | ) 167 | 168 | def download_assets(self, **kwargs): 169 | for file in self.download_files: 170 | asset = os.path.realpath(os.path.join(self.assets, file)) 171 | if not os.path.exists(asset): 172 | os.makedirs(os.path.dirname(asset), exist_ok=True) 173 | LOG.info("Downloading %s", asset) 174 | download(self.download_url.format(file=file), asset + ".download") 175 | os.rename(asset + ".download", asset) 176 | 177 | @property 178 | def asset_files(self, **kwargs): 179 | result = [] 180 | for file in self.download_files: 181 | result.append(os.path.realpath(os.path.join(self.assets, file))) 182 | return result 183 | 184 | @cached_property 185 | def device(self): 186 | import torch 187 | 188 | device = "cpu" 189 | 190 | if torch.backends.mps.is_available() and torch.backends.mps.is_built(): 191 | device = "mps" 192 | 193 | if torch.cuda.is_available() and torch.backends.cuda.is_built(): 194 | device = "cuda" 195 | 196 | LOG.info( 197 | "Using device '%s'. The speed of inference depends greatly on the device.", 198 | device.upper(), 199 | ) 200 | 201 | if self.only_gpu: 202 | if device == "cpu": 203 | raise RuntimeError("GPU is not available") 204 | 205 | return device 206 | 207 | def torch_deterministic_mode(self): 208 | import torch 209 | 210 | LOG.info("Setting deterministic mode for PyTorch") 211 | os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8" 212 | 213 | torch.backends.cudnn.benchmark = False 214 | torch.backends.cudnn.deterministic = True 215 | torch.use_deterministic_algorithms(True) 216 | 217 | @cached_property 218 | def providers(self): 219 | import onnxruntime as ort 220 | 221 | available_providers = ort.get_available_providers() 222 | providers = [] 223 | for n in ["CUDAExecutionProvider", "CPUExecutionProvider"]: 224 | if n in available_providers: 225 | providers.append(n) 226 | 227 | LOG.info( 228 | "Using device '%s'. The speed of inference depends greatly on the device.", 229 | ort.get_device(), 230 | ) 231 | 232 | if self.only_gpu: 233 | assert isinstance(ort.get_device(), str) 234 | if ort.get_device() == "CPU": 235 | raise RuntimeError("GPU is not available") 236 | 237 | providers = ["CUDAExecutionProvider"] 238 | 239 | LOG.info("ONNXRuntime providers: %s", providers) 240 | 241 | return providers 242 | 243 | def timer(self, title): 244 | return Timer(title) 245 | 246 | def stepper(self, step): 247 | # We assume that we call this method only once 248 | # just before the first iteration. 249 | elapsed = time.time() - self.created 250 | LOG.info("Model initialisation: %s", seconds(elapsed)) 251 | return Stepper(step, self.lead_time) 252 | 253 | def _datetimes(self, dates): 254 | date = self.date 255 | assert isinstance(date, int) 256 | if date <= 0: 257 | date = datetime.datetime.utcnow() + datetime.timedelta(days=date) 258 | date = date.year * 10000 + date.month * 100 + date.day 259 | 260 | time = self.time 261 | assert isinstance(time, int) 262 | if time < 100: 263 | time *= 100 264 | # assert time in (0, 600, 1200, 1800), time 265 | 266 | lagged = self.lagged 267 | if not lagged: 268 | lagged = [0] 269 | 270 | result = [] 271 | for basedate in dates: 272 | for lag in lagged: 273 | date = basedate + datetime.timedelta(hours=lag) 274 | result.append( 275 | ( 276 | date.year * 10000 + date.month * 100 + date.day, 277 | date.hour, 278 | ), 279 | ) 280 | 281 | return result 282 | 283 | def datetimes(self, step=0): 284 | if self.staging_dates: 285 | assert step == 0, step 286 | dates = [] 287 | with open(self.staging_dates) as f: 288 | for line in f: 289 | dates.append(datetime.datetime.fromisoformat(line.strip())) 290 | 291 | return self._datetimes(dates) 292 | 293 | date = self.date 294 | assert isinstance(date, int) 295 | if date <= 0: 296 | date = datetime.datetime.utcnow() + datetime.timedelta(days=date) 297 | date = date.year * 10000 + date.month * 100 + date.day 298 | 299 | time = self.time 300 | assert isinstance(time, int) 301 | if time < 100: 302 | time *= 100 303 | 304 | # assert time in (0, 600, 1200, 1800), time 305 | 306 | full = datetime.datetime( 307 | date // 10000, 308 | date % 10000 // 100, 309 | date % 100, 310 | time // 100, 311 | time % 100, 312 | ) + datetime.timedelta(hours=step) 313 | return self._datetimes([full]) 314 | 315 | def print_fields(self): 316 | param, level = self.param_level_pl 317 | print("Grid:", self.grid) 318 | print("Area:", self.area) 319 | print("Pressure levels:") 320 | print(" Levels:", level) 321 | print(" Params:", param) 322 | print("Single levels:") 323 | print(" Params:", self.param_sfc) 324 | 325 | def print_assets_list(self): 326 | for file in self.download_files: 327 | print(file) 328 | 329 | def _print_request(self, verb, request, file=sys.stdout): 330 | r = [verb] 331 | for k, v in sorted(request.items()): 332 | if not isinstance(v, (list, tuple, set)): 333 | v = [v] 334 | 335 | if k in ("area", "grid", "frame", "rotation", "bitmap"): 336 | v = [str(_) for _ in v] 337 | else: 338 | v = [str(_) for _ in sorted(v)] 339 | 340 | v = "/".join(v) 341 | r.append(f"{k}={v}") 342 | 343 | r = ",\n ".join(r) 344 | print(r, file=file) 345 | print(file=file) 346 | 347 | @property 348 | def _requests_extra(self): 349 | if not self.requests_extra: 350 | return {} 351 | extra = [_.split("=") for _ in self.requests_extra.split(",")] 352 | extra = {a: b for a, b in extra} 353 | return extra 354 | 355 | def print_requests(self): 356 | requests = self._requests() 357 | 358 | if self.json: 359 | print(json.dumps(requests, indent=4)) 360 | return 361 | 362 | for r in requests: 363 | self._print_request("retrieve", r) 364 | 365 | def _requests_unfiltered(self): 366 | result = [] 367 | 368 | first = dict( 369 | target="input.grib", 370 | grid=self.grid, 371 | area=self.area, 372 | ) 373 | first.update(self.retrieve) 374 | 375 | for date, time in self.datetimes(): # noqa F402 376 | param, level = self.param_level_pl 377 | 378 | r = dict( 379 | date=date, 380 | time=time, 381 | ) 382 | r.update(first) 383 | first = {} 384 | 385 | if param and level: 386 | 387 | # PL 388 | r.update( 389 | dict( 390 | levtype="pl", 391 | levelist=level, 392 | param=param, 393 | date=date, 394 | time=time, 395 | ) 396 | ) 397 | 398 | r.update(self._requests_extra) 399 | 400 | self.patch_retrieve_request(r) 401 | 402 | result.append(dict(**r)) 403 | 404 | # ML 405 | param, level = self.param_level_ml 406 | 407 | if param and level: 408 | r.update( 409 | dict( 410 | levtype="ml", 411 | levelist=level, 412 | param=param, 413 | date=date, 414 | time=time, 415 | ) 416 | ) 417 | 418 | r.update(self._requests_extra) 419 | 420 | self.patch_retrieve_request(r) 421 | 422 | result.append(dict(**r)) 423 | 424 | param = self.param_sfc 425 | if param: 426 | # SFC 427 | r.update( 428 | dict( 429 | levtype="sfc", 430 | param=self.param_sfc, 431 | date=date, 432 | time=time, 433 | levelist="off", 434 | ) 435 | ) 436 | 437 | self.patch_retrieve_request(r) 438 | result.append(dict(**r)) 439 | 440 | return result 441 | 442 | def _requests(self): 443 | 444 | def filter_constant(request): 445 | # We check for 'sfc' because param 'z' can be ambiguous 446 | if request.get("levtype") == "sfc": 447 | param = set(self.constant_fields) & set(request.get("param", [])) 448 | if param: 449 | request["param"] = list(param) 450 | return True 451 | 452 | return False 453 | 454 | def filter_prognostic(request): 455 | # TODO: We assume here that prognostic fields are 456 | # the ones that are not constant. This may not always be true 457 | if request.get("levtype") == "sfc": 458 | param = set(request.get("param", [])) - set(self.constant_fields) 459 | if param: 460 | request["param"] = list(param) 461 | return True 462 | return False 463 | 464 | return True 465 | 466 | def filter_last_date(request): 467 | date, time = max(self.datetimes()) 468 | return request["date"] == date and request["time"] == time 469 | 470 | def noop(request): 471 | return request 472 | 473 | filter_type = { 474 | "constants": filter_constant, 475 | "prognostics": filter_prognostic, 476 | "all": noop, 477 | }[self.retrieve_fields_type] 478 | 479 | filter_dates = { 480 | True: filter_last_date, 481 | False: noop, 482 | }[self.retrieve_only_one_date] 483 | 484 | result = [] 485 | for r in self._requests_unfiltered(): 486 | if filter_type(r) and filter_dates(r): 487 | result.append(r) 488 | 489 | return result 490 | 491 | def patch_retrieve_request(self, request): 492 | # Overriden in subclasses if needed 493 | pass 494 | 495 | def peek_into_checkpoint(self, path): 496 | return peek(path) 497 | 498 | def parse_model_args(self, args): 499 | if args: 500 | raise NotImplementedError(f"This model does not accept arguments {args}") 501 | 502 | def provenance(self): 503 | from .provenance import gather_provenance_info 504 | 505 | return gather_provenance_info(self.asset_files) 506 | 507 | def forcing_and_constants(self, date, param): 508 | source = self.all_fields[:1] 509 | 510 | ds = ekd.from_source( 511 | "forcings", 512 | source, 513 | date=date, 514 | param=param, 515 | ) 516 | 517 | assert len(ds) == len(param), (len(ds), len(param), date) 518 | 519 | return ds.to_numpy(dtype=np.float32) 520 | 521 | @cached_property 522 | def gridpoints(self): 523 | return len(self.all_fields[0].grid_points()[0]) 524 | 525 | @cached_property 526 | def start_datetime(self): 527 | return self.all_fields.order_by(valid_datetime="ascending")[-1].datetime()["valid_time"] 528 | 529 | @property 530 | def constant_fields(self): 531 | raise NotImplementedError("constant_fields") 532 | 533 | def write_input_fields( 534 | self, 535 | fields, 536 | accumulations=None, 537 | accumulations_template=None, 538 | accumulations_shape=None, 539 | ignore=None, 540 | ): 541 | LOG.info("Starting date is %s", self.start_datetime) 542 | LOG.info("Writing input fields") 543 | if ignore is None: 544 | ignore = [] 545 | 546 | with self.timer("Writing step 0"): 547 | for field in fields: 548 | if field.metadata("shortName") in ignore: 549 | continue 550 | 551 | if field.datetime()["valid_time"] == self.start_datetime: 552 | self.write( 553 | None, 554 | template=field, 555 | step=0, 556 | ) 557 | 558 | if accumulations is not None: 559 | if accumulations_template is None: 560 | accumulations_template = fields.sel(param="msl")[0] 561 | 562 | if accumulations_shape is None: 563 | accumulations_shape = accumulations_template.shape 564 | 565 | if accumulations_template.metadata("edition") == 1: 566 | for param in accumulations: 567 | 568 | self.write( 569 | np.zeros(accumulations_shape, dtype=np.float32), 570 | stepType="accum", 571 | template=accumulations_template, 572 | param=param, 573 | startStep=0, 574 | endStep=0, 575 | date=int(self.start_datetime.strftime("%Y%m%d")), 576 | time=int(self.start_datetime.strftime("%H%M")), 577 | check=True, 578 | ) 579 | else: 580 | # # TODO: Remove this when accumulations are supported for GRIB edition 2 581 | 582 | template = """ 583 | R1JJQv//AAIAAAAAAAAA3AAAABUBAGIAABsBAQfoCRYGAAAAAQAAABECAAEAAQAJBAIwMDAxAAAA 584 | SAMAAA/XoAAAAAAG////////////////////AAAFoAAAAtEAAAAA/////wVdSoAAAAAAMIVdSoAV 585 | cVlwAAPQkAAD0JAAAAAAOgQAAAAIAcEC//8AAAABAAAAAAH//////////////wfoCRYGAAABAAAA 586 | AAECAQAAAAD/AAAAAAAAABUFAA/XoAAAAAAAAIAKAAAAAAAAAAYG/wAAAAUHNzc3N0dSSUL//wAC 587 | AAAAAAAAANwAAAAVAQBiAAAbAQEH6AkWDAAAAAEAAAARAgABAAEACQQBMDAwMQAAAEgDAAAP16AA 588 | AAAABv///////////////////wAABaAAAALRAAAAAP////8FXUqAAAAAADCFXUqAFXFZcAAD0JAA 589 | A9CQAAAAADoEAAAACAHBAv//AAAAAQAAAAAB//////////////8H6AkWDAAAAQAAAAABAgEAAAAA 590 | /wAAAAAAAAAVBQAP16AAAAAAAACACgAAAAAAAAAGBv8AAAAFBzc3Nzc= 591 | """ 592 | 593 | template = base64.b64decode(template) 594 | accumulations_template = ekd.from_source("memory", template)[0] 595 | 596 | for param in accumulations: 597 | self.write( 598 | np.zeros(accumulations_shape, dtype=np.float32), 599 | stepType="accum", 600 | template=accumulations_template, 601 | param=param, 602 | startStep=0, 603 | endStep=0, 604 | date=int(self.start_datetime.strftime("%Y%m%d")), 605 | time=int(self.start_datetime.strftime("%H%M")), 606 | check=True, 607 | ) 608 | 609 | 610 | def load_model(name, **kwargs): 611 | return available_models()[name].load()(**kwargs) 612 | 613 | 614 | def available_models(): 615 | result = {} 616 | for e in entrypoints.get_group_all("ai_models.model"): 617 | result[e.name] = e 618 | return result 619 | -------------------------------------------------------------------------------- /src/ai_models/outputs/__init__.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import itertools 9 | import logging 10 | import warnings 11 | from functools import cached_property 12 | 13 | import earthkit.data as ekd 14 | import entrypoints 15 | import numpy as np 16 | 17 | LOG = logging.getLogger(__name__) 18 | 19 | 20 | class Output: 21 | def write(self, *args, **kwargs): 22 | pass 23 | 24 | def flush(self, *args, **kwargs): 25 | pass 26 | 27 | 28 | class GribOutputBase(Output): 29 | def __init__(self, owner, path, metadata, **kwargs): 30 | self._first = True 31 | metadata.setdefault("stream", "oper") 32 | metadata.setdefault("expver", owner.expver) 33 | metadata.setdefault("class", "ml") 34 | 35 | self.path = path 36 | self.owner = owner 37 | self.metadata = metadata 38 | 39 | @cached_property 40 | def grib_keys(self): 41 | edition = self.metadata.pop("edition", self.owner.grib_edition) 42 | 43 | _grib_keys = dict( 44 | edition=edition, 45 | generatingProcessIdentifier=self.owner.version, 46 | ) 47 | _grib_keys.update(self.metadata) 48 | 49 | return _grib_keys 50 | 51 | @cached_property 52 | def output(self): 53 | return ekd.new_grib_output( 54 | self.path, 55 | split_output=True, 56 | **self.grib_keys, 57 | ) 58 | 59 | def write(self, data, *args, check=False, **kwargs): 60 | 61 | try: 62 | handle, path = self.output.write(data, *args, **kwargs) 63 | 64 | except Exception: 65 | if data is not None: 66 | if np.isnan(data).any(): 67 | raise ValueError(f"NaN values found in field. args={args} kwargs={kwargs}") 68 | if np.isinf(data).any(): 69 | raise ValueError(f"Infinite values found in field. args={args} kwargs={kwargs}") 70 | 71 | options = {} 72 | options.update(self.grib_keys) 73 | options.update(kwargs) 74 | LOG.error("Failed to write data to %s %s", args, options) 75 | cmd = [] 76 | for k, v in options.items(): 77 | if isinstance(v, (int, str, float)): 78 | cmd.append("%s=%s" % (k, v)) 79 | 80 | LOG.error("grib_set -s%s", ",".join(cmd)) 81 | 82 | raise 83 | 84 | if check: 85 | # Check that the GRIB keys are as expected 86 | 87 | if kwargs.get("expver") is None: 88 | ignore = ("template", "check_nans", "expver", "class", "type", "stream") 89 | else: 90 | ignore = ("template", "check_nans") 91 | 92 | for key, value in itertools.chain(self.grib_keys.items(), kwargs.items()): 93 | if key in ignore: 94 | continue 95 | 96 | # If "param" is a string, we what to compare it to the shortName 97 | if key == "param": 98 | try: 99 | float(value) 100 | except ValueError: 101 | key = "shortName" 102 | 103 | assert str(handle.get(key)) == str(value), (key, handle.get(key), value) 104 | 105 | return handle, path 106 | 107 | 108 | class FileOutput(GribOutputBase): 109 | def __init__(self, *args, **kwargs): 110 | super().__init__(*args, **kwargs) 111 | LOG.info("Writing results to %s", self.path) 112 | 113 | 114 | class NoneOutput(Output): 115 | def __init__(self, *args, **kwargs): 116 | LOG.info("Results will not be written.") 117 | 118 | def write(self, *args, **kwargs): 119 | pass 120 | 121 | 122 | class HindcastReLabel: 123 | def __init__(self, owner, output, hindcast_reference_year=None, hindcast_reference_date=None, **kwargs): 124 | self.owner = owner 125 | self.output = output 126 | self.hindcast_reference_year = int(hindcast_reference_year) if hindcast_reference_year else None 127 | self.hindcast_reference_date = int(hindcast_reference_date) if hindcast_reference_date else None 128 | assert self.hindcast_reference_year is not None or self.hindcast_reference_date is not None 129 | 130 | def write(self, *args, **kwargs): 131 | if "hdate" in kwargs: 132 | warnings.warn(f"Ignoring hdate='{kwargs['hdate']}' in HindcastReLabel", stacklevel=3) 133 | kwargs.pop("hdate") 134 | 135 | if "date" in kwargs: 136 | warnings.warn(f"Ignoring date='{kwargs['date']}' in HindcastReLabel", stacklevel=3) 137 | kwargs.pop("date") 138 | 139 | date = kwargs["template"]["date"] 140 | hdate = kwargs["template"]["hdate"] 141 | 142 | if hdate is not None: 143 | # Input was a hindcast 144 | referenceDate = ( 145 | self.hindcast_reference_date 146 | if self.hindcast_reference_date is not None 147 | else self.hindcast_reference_year * 10000 + date % 10000 148 | ) 149 | assert date == referenceDate, ( 150 | date, 151 | referenceDate, 152 | hdate, 153 | kwargs["template"], 154 | ) 155 | kwargs["referenceDate"] = referenceDate 156 | kwargs["hdate"] = hdate 157 | else: 158 | referenceDate = ( 159 | self.hindcast_reference_date 160 | if self.hindcast_reference_date is not None 161 | else self.hindcast_reference_year * 10000 + date % 10000 162 | ) 163 | kwargs["referenceDate"] = referenceDate 164 | kwargs["hdate"] = date 165 | 166 | kwargs.setdefault("check", True) 167 | 168 | return self.output.write(*args, **kwargs) 169 | 170 | def flush(self, *args, **kwargs): 171 | return self.output.flush(*args, **kwargs) 172 | 173 | 174 | class NoLabelling: 175 | 176 | def __init__(self, owner, output, **kwargs): 177 | self.owner = owner 178 | self.output = output 179 | 180 | def write(self, *args, **kwargs): 181 | kwargs["deleteLocalDefinition"] = 1 182 | return self.output.write(*args, **kwargs) 183 | 184 | def flush(self, *args, **kwargs): 185 | return self.output.flush(*args, **kwargs) 186 | 187 | 188 | def get_output(name, owner, *args, **kwargs): 189 | result = available_outputs()[name].load()(owner, *args, **kwargs) 190 | if kwargs.get("hindcast_reference_year") is not None or kwargs.get("hindcast_reference_date") is not None: 191 | result = HindcastReLabel(owner, result, **kwargs) 192 | if owner.expver is None: 193 | result = NoLabelling(owner, result, **kwargs) 194 | return result 195 | 196 | 197 | def available_outputs(): 198 | result = {} 199 | for e in entrypoints.get_group_all("ai_models.output"): 200 | result[e.name] = e 201 | return result 202 | -------------------------------------------------------------------------------- /src/ai_models/remote/__init__.py: -------------------------------------------------------------------------------- 1 | from .api import RemoteAPI 2 | from .model import RemoteModel 3 | 4 | __all__ = ["RemoteAPI", "RemoteModel"] 5 | -------------------------------------------------------------------------------- /src/ai_models/remote/api.py: -------------------------------------------------------------------------------- 1 | import logging 2 | import os 3 | import sys 4 | import time 5 | from urllib.parse import urljoin 6 | 7 | import requests 8 | from multiurl import download 9 | from multiurl import robust 10 | from tqdm import tqdm 11 | 12 | from .config import API_URL 13 | from .config import CONFIG_PATH 14 | from .config import load_config 15 | 16 | LOG = logging.getLogger(__name__) 17 | 18 | 19 | class BearerAuth(requests.auth.AuthBase): 20 | def __init__(self, token): 21 | self.token = token 22 | 23 | def __call__(self, r): 24 | r.headers["authorization"] = "Bearer " + self.token 25 | return r 26 | 27 | 28 | class RemoteAPI: 29 | def __init__( 30 | self, 31 | input_file: str = None, 32 | output_file: str = "output.grib", 33 | url: str = None, 34 | token: str = None, 35 | ): 36 | config = load_config() 37 | 38 | self.url = url or os.getenv("AI_MODELS_REMOTE_URL") or config.get("url") or API_URL 39 | if not self.url.endswith("/"): 40 | self.url += "/" 41 | 42 | self.token = token or os.getenv("AI_MODELS_REMOTE_TOKEN") or config.get("token") 43 | 44 | if not self.token: 45 | LOG.error( 46 | "Missing remote token. Set it in %s or env AI_MODELS_REMOTE_TOKEN", 47 | CONFIG_PATH, 48 | ) 49 | sys.exit(1) 50 | 51 | LOG.info("Using remote server %s", self.url) 52 | 53 | self.auth = BearerAuth(self.token) 54 | self.output_file = output_file 55 | self.input_file = input_file 56 | self._timeout = 300 57 | 58 | def run(self, cfg: dict): 59 | # upload file 60 | with open(self.input_file, "rb") as file: 61 | LOG.info("Uploading input file to remote server") 62 | data = self._request(requests.post, "upload", data=file) 63 | 64 | if data["status"] != "success": 65 | LOG.error(data["status"]) 66 | if reason := data.get("reason"): 67 | LOG.error(reason) 68 | sys.exit(1) 69 | 70 | # submit task 71 | data = self._request(requests.post, data["href"], json=cfg) 72 | 73 | LOG.info("Inference request submitted") 74 | 75 | if data["status"] != "queued": 76 | LOG.error(data["status"]) 77 | if reason := data.get("reason"): 78 | LOG.error(reason) 79 | sys.exit(1) 80 | 81 | LOG.info("Request id: %s", data["id"]) 82 | LOG.info("Request is queued") 83 | 84 | last_status = data["status"] 85 | pbar = None 86 | 87 | while True: 88 | data = self._request(requests.get, data["href"]) 89 | 90 | if data["status"] == "ready": 91 | if pbar is not None: 92 | pbar.close() 93 | LOG.info("Request is ready") 94 | break 95 | 96 | if data["status"] == "failed": 97 | LOG.error("Request failed") 98 | if reason := data.get("reason"): 99 | LOG.error(reason) 100 | sys.exit(1) 101 | 102 | if data["status"] != last_status: 103 | LOG.info("Request is %s", data["status"]) 104 | last_status = data["status"] 105 | 106 | if progress := data.get("progress"): 107 | if pbar is None: 108 | pbar = tqdm( 109 | total=progress.get("total", 0), 110 | unit="steps", 111 | ncols=70, 112 | leave=False, 113 | initial=1, 114 | bar_format="{desc}: {percentage:3.0f}%|{bar}| {n_fmt}/{total_fmt} {unit}{postfix}", 115 | ) 116 | if eta := progress.get("eta"): 117 | pbar.set_postfix_str(f"ETA: {eta}") 118 | if status := progress.get("status"): 119 | pbar.set_description(status.strip().capitalize()) 120 | pbar.update(progress.get("step", 0) - pbar.n) 121 | 122 | time.sleep(5) 123 | 124 | download(urljoin(self.url, data["href"]), target=self.output_file) 125 | 126 | LOG.debug("Result written to %s", self.output_file) 127 | 128 | def metadata(self, model, model_version, param) -> dict: 129 | if isinstance(param, str): 130 | return self._request(requests.get, f"metadata/{model}/{model_version}/{param}") 131 | elif isinstance(param, (list, dict)): 132 | return self._request(requests.post, f"metadata/{model}/{model_version}", json=param) 133 | else: 134 | raise ValueError("param must be a string, list, or dict with 'param' key.") 135 | 136 | def models(self): 137 | results = self._request(requests.get, "models") 138 | 139 | if not isinstance(results, list): 140 | return [] 141 | 142 | return results 143 | 144 | def patch_retrieve_request(self, cfg, request): 145 | cfg["patchrequest"] = request 146 | result = self._request(requests.post, "patch", json=cfg) 147 | if status := result.get("status"): 148 | LOG.error(status) 149 | sys.exit(1) 150 | return result 151 | 152 | def _request(self, type, href, data=None, json=None, auth=None): 153 | response = robust(type, retry_after=30)( 154 | urljoin(self.url, href), 155 | json=json, 156 | data=data, 157 | auth=self.auth, 158 | timeout=self._timeout, 159 | ) 160 | 161 | if response.status_code == 401: 162 | LOG.error("Unauthorized Access. Check your token.") 163 | sys.exit(1) 164 | 165 | try: 166 | data = response.json() 167 | 168 | if isinstance(data, dict) and (status := data.get("status")): 169 | data["status"] = status.lower() 170 | 171 | return data 172 | except Exception: 173 | return {"status": f"{response.url} {response.status_code} {response.text}"} 174 | -------------------------------------------------------------------------------- /src/ai_models/remote/config.py: -------------------------------------------------------------------------------- 1 | import logging 2 | import os 3 | 4 | API_URL = "https://ai-models.ecmwf.int/api/v1/" 5 | 6 | ROOT_PATH = os.path.join(os.path.expanduser("~"), ".config", "ai-models") 7 | CONFIG_PATH = os.path.join(ROOT_PATH, "api.yaml") 8 | 9 | LOG = logging.getLogger(__name__) 10 | 11 | 12 | def config_exists(): 13 | return os.path.exists(CONFIG_PATH) 14 | 15 | 16 | def create_config(): 17 | if config_exists(): 18 | return 19 | 20 | try: 21 | os.makedirs(ROOT_PATH, exist_ok=True) 22 | with open(CONFIG_PATH, "w") as f: 23 | f.write("token: \n") 24 | f.write(f"url: {API_URL}\n") 25 | except Exception as e: 26 | LOG.error(f"Failed to create config {CONFIG_PATH}") 27 | LOG.error(e, exc_info=True) 28 | 29 | 30 | def load_config() -> dict: 31 | from yaml import safe_load 32 | 33 | if not config_exists(): 34 | create_config() 35 | 36 | try: 37 | with open(CONFIG_PATH, "r") as f: 38 | return safe_load(f) or {} 39 | except Exception as e: 40 | LOG.error(f"Failed to read config {CONFIG_PATH}") 41 | LOG.error(e, exc_info=True) 42 | return {} 43 | -------------------------------------------------------------------------------- /src/ai_models/remote/model.py: -------------------------------------------------------------------------------- 1 | import logging 2 | import os 3 | import sys 4 | import tempfile 5 | from functools import cached_property 6 | 7 | import earthkit.data as ekd 8 | 9 | from ..model import Model 10 | from .api import RemoteAPI 11 | 12 | LOG = logging.getLogger(__name__) 13 | 14 | 15 | class RemoteModel(Model): 16 | def __init__(self, **kwargs): 17 | self.cfg = kwargs 18 | self.cfg["download_assets"] = False 19 | 20 | self.model = self.cfg["model"] 21 | self.model_version = self.cfg.get("model_version", "latest") 22 | self._param = {} 23 | self.api = RemoteAPI() 24 | 25 | if self.model not in self.api.models(): 26 | LOG.error(f"Model '{self.model}' not available on remote server.") 27 | LOG.error("Rerun the command with --models --remote to list available remote models.") 28 | sys.exit(1) 29 | 30 | self.load_parameters() 31 | 32 | super().__init__(**self.cfg) 33 | 34 | def __getattr__(self, name): 35 | return self.get_parameter(name) 36 | 37 | def run(self): 38 | with tempfile.TemporaryDirectory() as tmpdirname: 39 | input_file = os.path.join(tmpdirname, "input.grib") 40 | output_file = os.path.join(tmpdirname, "output.grib") 41 | self.all_fields.save(input_file) 42 | 43 | self.api.input_file = input_file 44 | self.api.output_file = output_file 45 | 46 | self.api.run(self.cfg) 47 | 48 | ds = ekd.from_source("file", output_file) 49 | for field in ds: 50 | self.write(None, template=field) 51 | 52 | def parse_model_args(self, args): 53 | return None 54 | 55 | def patch_retrieve_request(self, request): 56 | if not self.remote_has_patch: 57 | return 58 | 59 | patched = self.api.patch_retrieve_request(self.cfg, request) 60 | request.update(patched) 61 | 62 | def load_parameters(self): 63 | params = self.api.metadata( 64 | self.model, 65 | self.model_version, 66 | [ 67 | "expver", 68 | "version", 69 | "grid", 70 | "area", 71 | "param_level_ml", 72 | "param_level_pl", 73 | "param_sfc", 74 | "lagged", 75 | "grib_extra_metadata", 76 | "retrieve", 77 | "remote_has_patch", # custom parameter, checks if remote model need patches 78 | ], 79 | ) 80 | self._param.update(params) 81 | 82 | def get_parameter(self, name): 83 | if (param := self._param.get(name)) is not None: 84 | return param 85 | 86 | _param = self.api.metadata(self.model, self.model_version, name) 87 | self._param.update(_param) 88 | 89 | return self._param.get(name) 90 | 91 | @cached_property 92 | def param_level_ml(self): 93 | return self.get_parameter("param_level_ml") or ([], []) 94 | 95 | @cached_property 96 | def param_level_pl(self): 97 | return self.get_parameter("param_level_pl") or ([], []) 98 | 99 | @cached_property 100 | def param_sfc(self): 101 | return self.get_parameter("param_sfc") or [] 102 | 103 | @cached_property 104 | def lagged(self): 105 | return self.get_parameter("lagged") or False 106 | 107 | @cached_property 108 | def version(self): 109 | return self.get_parameter("version") or 1 110 | 111 | @cached_property 112 | def grib_extra_metadata(self): 113 | return self.get_parameter("grib_extra_metadata") or {} 114 | 115 | @cached_property 116 | def retrieve(self): 117 | return self.get_parameter("retrieve") or {} 118 | -------------------------------------------------------------------------------- /src/ai_models/stepper.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | import logging 9 | import time 10 | 11 | from earthkit.data.utils.humanize import seconds 12 | 13 | LOG = logging.getLogger(__name__) 14 | 15 | 16 | class Stepper: 17 | def __init__(self, step, lead_time): 18 | self.step = step 19 | self.lead_time = lead_time 20 | self.start = time.time() 21 | self.last = self.start 22 | self.num_steps = lead_time // step 23 | LOG.info("Starting inference for %s steps (%sh).", self.num_steps, lead_time) 24 | 25 | def __enter__(self): 26 | return self 27 | 28 | def __call__(self, i, step): 29 | now = time.time() 30 | elapsed = now - self.start 31 | speed = (i + 1) / elapsed 32 | eta = (self.num_steps - i) / speed 33 | LOG.info( 34 | "Done %s out of %s in %s (%sh), ETA: %s.", 35 | i + 1, 36 | self.num_steps, 37 | seconds(now - self.last), 38 | step, 39 | seconds(eta), 40 | ) 41 | self.last = now 42 | 43 | def __exit__(self, *args): 44 | if self.num_steps == 0: 45 | return 46 | 47 | elapsed = time.time() - self.start 48 | LOG.info("Elapsed: %s.", seconds(elapsed)) 49 | LOG.info("Average: %s per step.", seconds(elapsed / self.num_steps)) 50 | -------------------------------------------------------------------------------- /tests/requirements.txt: -------------------------------------------------------------------------------- 1 | # Empty for now 2 | -------------------------------------------------------------------------------- /tests/test_code.py: -------------------------------------------------------------------------------- 1 | # (C) Copyright 2023 European Centre for Medium-Range Weather Forecasts. 2 | # This software is licensed under the terms of the Apache Licence Version 2.0 3 | # which can be obtained at http://www.apache.org/licenses/LICENSE-2.0. 4 | # In applying this licence, ECMWF does not waive the privileges and immunities 5 | # granted to it by virtue of its status as an intergovernmental organisation 6 | # nor does it submit to any jurisdiction. 7 | 8 | 9 | def test_code(): 10 | pass # Empty for now 11 | --------------------------------------------------------------------------------