Add metrics calculation app

Dockerfile

@@ -37,6 +37,7 @@ FROM build-deps AS build-app
 # * .git: Required for setuptools-git-versioning
 COPY src /opt/app/src
 COPY .git /opt/app/.git
+COPY uv.lock /opt/app/uv.lock
 RUN uv sync --no-dev --no-editable --index-strategy unsafe-best-match
 
 # --- Runtime image (use distroless if feasible for 100MB saving) --- #
@@ -49,5 +50,24 @@ COPY --from=build-app --chown=app:app /opt/app/.venv /opt/app/.venv
 
 ENV _TYPER_STANDARD_TRACEBACK=1
 
-ENTRYPOINT ["/opt/app/.venv/bin/cloudcasting-app"]
+COPY <<EOF /opt/app/entrypoint.sh
+#!/bin/bash
+set -e
+
+case "\$1" in
+inference)
+    /opt/app/.venv/bin/cloudcasting-inference
+    ;;
+metrics)
+    /opt/app/.venv/bin/cloudcasting-metrics
+    ;;
+*)
+    exit 1
+    ;;
+esac
+EOF
+
+RUN chmod +x /opt/app/entrypoint.sh
+
+ENTRYPOINT ["/opt/app/entrypoint.sh"]
 

README.md

@@ -6,56 +6,42 @@
 [![contributors badge](https://img.shields.io/github/contributors/openclimatefix/cloudcasting-app?color=FFFFFF)](https://github.com/openclimatefix/cloudcasting-app/graphs/contributors)
 [![ease of contribution: hard](https://img.shields.io/badge/ease%20of%20contribution:%20hard-bb2629)](https://github.com/openclimatefix#how-easy-is-it-to-get-involved)
 
-This repo is used to run the OCF-ATI cloudcasting model live in production. This model takes
-previous frames of EUMETSAT satellite images and forecasts the future frames to come.
+This repo is used to run the OCF-ATI cloudcasting model live in production and meausure its 
+performance. The model takes previous frames of EUMETSAT satellite images and forecasts the future 
+frames to come.
 
-The repo associated with training the models run here is https://github.com/openclimatefix/sat_pred
-
-The model checkpoints are hosted at:
-https://huggingface.co/openclimatefix/cloudcasting_uk
-
-## Environment Variables
-
-The following environment variables are used in the app:
-
-- `SATELLITE_ZARR_PATH`: The path to the satellite data in Zarr format.
-- `OUTPUT_PREDICTION_DIRECTORY`: The directory where results are saved. 
-
-### Optional Environment Variables
-
-- `SATELLITE_SCALE_FACTOR`: The scale factor for the satellite data. Defaults to 1023. 
-- `SATELLITE_15_ZARR_PATH`: The path to the 15 minute satellite data in Zarr format. If 
-this is not set then the `SATELLITE_ZARR_PATH` is used by `.zarr` is repalced with `_15.zarr`
+This repo contains two different packages:
+ - `cloudcasting_inference`: Used to run inference
+ - `cloudcasting_metrics`: Used to score the predictions against ground truth
 
 ## Installation
 
 ## Setup / Installation
 
+Both packages will be installed simultaneously using:
+
 ```bash
 git clone https://github.com/openclimatefix/cloudcasting-app
 cd cloudcasting-app
 pip install .
 ```
 
+## Usage and environmental variables
 
-## Example usage
-
-### Running the app locally
-
-It is possbile to run the app locally by setting the required environment variables listed at the top of the [app](src/cloudcasting_app/app.py), these should point to the relevant paths for loading satellite data 
-and saving predicitons.
+See the READMEs in `src/cloudcasting_inference` and `src/cloudcasting_metrics`.
 
 ## Development
 
 ### Running the test suite
 
-The test suite is via pytest and can be run from command line using
+The test suite is via pytest and can be run from command line using:
 
 ```
 pytest
 ```
-
 
+This will run tests for both packages.
+
 ## Contributing and community
 
 [![issues badge](https://img.shields.io/github/issues/openclimatefix/cloudcasting-app?color=FFAC5F)](https://github.com/openclimatefix/cloudcasting-app/issues?q=is%3Aissue+is%3Aopen+sort%3Aupdated-desc)
@@ -65,7 +51,6 @@ pytest
 - Check out the [OCF blog](https://openclimatefix.org/blog) for updates
 - Follow OCF on [LinkedIn](https://uk.linkedin.com/company/open-climate-fix)
 
-
 ## Contributors
 
 <!-- ALL-CONTRIBUTORS-LIST:START - Do not remove or modify this section -->

pyproject.toml

@@ -21,23 +21,23 @@ classifiers = [
 ]
 dependencies = [
     "fsspec==2025.7.0",
-    "huggingface-hub==0.28.1",
-    "hydra-core==1.3.2",
-    "loguru == 0.7.3",
-    "numpy==2.1.2",
+    "huggingface-hub",
+    "hydra-core",
+    "icechunk",
+    "loguru",
+    "numpy",
     "ocf-data-sampler==0.5.27",
-    "pandas==2.2.3",
-    "s3fs==2025.7.0",
-    "safetensors==0.5.2",
+    "pandas",
+    "s3fs",
+    "safetensors",
     "sat_pred @ git+https://github.com/openclimatefix/sat_pred.git@main",
     # Since torch distributes CPU only packages as wheels, have to specify the target platform in order to pull the wheel compiled for that specific platform
     "torch @ https://download.pytorch.org/whl/cpu/torch-2.3.1-cp312-none-macosx_11_0_arm64.whl ; platform_system == 'Darwin' and platform_machine == 'arm64'",
     "torchvision @ https://download.pytorch.org/whl/cpu/torchvision-0.18.1-cp312-cp312-macosx_11_0_arm64.whl ; platform_system == 'Darwin' and platform_machine == 'arm64'",
     "torch @ https://download.pytorch.org/whl/cpu/torch-2.3.1%2Bcpu-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64'",
     "torchvision @ https://download.pytorch.org/whl/cpu/torchvision-0.18.1%2Bcpu-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64'",
-    "typer==0.15.1",
-    "xarray==2025.7.1",
-    "zarr==3.1.1",
+    "xarray",
+    "zarr",
 ]
 
 [dependency-groups]
@@ -52,11 +52,15 @@ dev = [
 
 [project.scripts]
 # Put entrypoints in here
-cloudcasting-app = "cloudcasting_app.app:main"
+cloudcasting-inference = "cloudcasting_inference.app:app"
+cloudcasting-metrics = "cloudcasting_metrics.app:app"
 
 [project.urls]
 repository = "https://github.com/openclimatefix/cloudcasting-app"
 
+[tool.setuptools.packages.find]
+where = ["src"]
+
 [tool.setuptools]
 include-package-data = false
 
@@ -122,18 +126,6 @@ line-ending = "auto"
 docstring-code-format = true
 docstring-code-line-length = 100
 
-# --- DOCUMENTATION CONFIGURATION --- #
-
-[tool.pydoctor]
-add-package = ["src/cloudcasting_app"]
-project-base-dir = "src/cloudcasting_app"
-docformat = "google"
-html-output = "docs"
-theme = "classic"
-privacy = [
-    "HIDDEN:**.test_*",
-]
-
 [[tool.uv.index]]
 url = "https://pypi.org/simple"
 

src/cloudcasting_inference/README.md

@@ -0,0 +1,31 @@
+# cloudcasting-inference
+
+This package is used to run inference on the model. This model takes previous frames of EUMETSAT 
+satellite images and forecasts the future frames to come.
+
+The repo associated with training the models run here is https://github.com/openclimatefix/sat_pred
+
+The model checkpoints are hosted at:
+https://huggingface.co/openclimatefix/cloudcasting_uk
+
+## Environment Variables
+
+The following environment variables are used in the app:
+
+- `SATELLITE_ZARR_PATH`: The path to the satellite data in Zarr format.
+- `PREDICTION_SAVE_DIRECTORY`: The directory where predictions will be saved. 
+
+### Optional Environment Variables
+
+- `SATELLITE_15_ZARR_PATH`: The path to the 15 minute satellite data in Zarr format. If 
+this is not set then the `SATELLITE_ZARR_PATH` is used by `.zarr` is repalced with `_15.zarr`
+
+## Example usage
+
+### Running the app locally
+
+It is possible to run the app locally by setting the required environment variables listed at the
+top of the [inference app](src/cloudcasting_inference/app.py), these should point to the relevant 
+paths for loading satellite data and saving predicitons.
+
+You can then run the app using `uv run cloudcasting-inference`.s

src/cloudcasting_app/__init__.py → src/cloudcasting_inference/__init__.py

@@ -43,9 +43,4 @@ def structured_formatter(record: "loguru.Record") -> str:
     loguru.logger.add(
         sys.stdout, format=structured_formatter, backtrace=True,
         level=os.getenv("LOGLEVEL", "INFO").upper(),
-    )
-
-# Uncomment and change the list to quieten external libraries
-# for logger in ["aiobotocore", "cfgrib"]:
-#    logging.getLogger(logger).setLevel(logging.WARNING)
-
+    )

src/cloudcasting_app/app.py → src/cloudcasting_inference/app.py

@@ -2,7 +2,7 @@
 
 This app expects these environmental variables to be available:
     SATELLITE_ZARR_PATH (str): The path of the input satellite data
-    OUTPUT_PREDICTION_DIRECTORY (str): The path of the directory to save the predictions to
+    PREDICTION_SAVE_DIRECTORY (str): The path of the directory to save the predictions to
 """
 
 import os
@@ -12,18 +12,17 @@
 import hydra
 import pandas as pd
 import torch
-import typer
 import xarray as xr
 import yaml
 from huggingface_hub import snapshot_download
 from safetensors.torch import load_model
 from loguru import logger
 
-from cloudcasting_app.data import SatelliteDownloader, sat_path, get_input_data
+from cloudcasting_inference.data import SatelliteDownloader, sat_path, get_input_data
 
 # Get package version
 try:
-    __version__ = version("cloudcasting-app")
+    __version__ = version("cloudcasting-inference")
 except PackageNotFoundError:
     __version__ = "v?"
 
@@ -33,20 +32,18 @@
 # Model will use GPU if available
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-
 # Model revision on huggingface
 REPO_ID = "openclimatefix/cloudcasting_uk"
 REVISION = "47643e89000e64e0150f7359ccc0cb6524948712"
 
 
-
 def app(t0=None):
     """Inference function for production
 
     Args:
         t0 (datetime): Datetime at which forecast is made
     """
-    logger.info(f"Using `cloudcasting_app` version: {__version__}", version=__version__)
+    logger.info(f"Using `cloudcasting-app` version: {__version__}", version=__version__)
 
     # ---------------------------------------------------------------------------
     # 0. If inference datetime is None, round down to last 30 minutes
@@ -123,7 +120,7 @@ def app(t0=None):
     ds_y_hat.sat_pred.attrs.update(ds.data.attrs)
 
     # Save predictions to the latest path and to path with timestring
-    out_dir = os.environ["OUTPUT_PREDICTION_DIRECTORY"]
+    out_dir = os.environ["PREDICTION_SAVE_DIRECTORY"]
 
     if satellite_downloader.use_5_minute:
         latest_zarr_path = f"{out_dir}/latest.zarr"
@@ -141,11 +138,3 @@ def app(t0=None):
             fs.rm(path, recursive=True)
 
         ds_y_hat.to_zarr(path)
-
-
-def main() -> None:
-    """Entrypoint to the application."""
-    typer.run(app)
-
-if __name__ == "__main__":
-    main()

src/cloudcasting_metrics/README.md

@@ -0,0 +1,14 @@
+# cloudcasting-metrics
+
+This package is used to score the predictions made by the cloudcasting model and save the scores to 
+a zarr store at the location `METRIC_ZARR_PATH`. By default, the scoring is run for all init-times 
+made the day before running.
+
+## Environment Variables
+
+The following environment variables are used in the app:
+
+- `SATELLITE_ICECHUNK_ARCHIVE`: The path to the satellite archive in icechunk formast
+- `PREDICTION_SAVE_DIRECTORY`: The directory where the cloudcasting predictions are saved. 
+  i.e. set to the same as `PREDICTION_SAVE_DIRECTORY` in `cloudcasting_inference`.
+- `METRIC_ZARR_PATH`: Where to save metrics zarr