altaris / acc23

Allergy Chip Challenge 2023

Allergy Chip Challenge 2023

Minimal example

from acc23 import models, ACCDataModule
from acc23.utils import train_model

# 1. Choose a model
model = models.Orchid()

# 2. Construct a datamodule from the competition data
datamodule = ACCDataModule("data/train.csv", "data/test.csv", "data/images")

# 3. Train the model. This can of course be done directly with pytorch
# lightning's API, or even a classic pytorch training loop
model = train_model(model, datamodule, root_dir="out")

# 4. Evaluate the model on the test dataset. The output file can readily
# be submitted to trustii.io!
df = eval_on_test_dataset(model, datamodule, root_dir="out/eval/test")
df.to_csv(f"out/predictions.csv", index=False)

Package organization

The important user-facing modules are:

• acc23.models: Subpackage that contains all model definitions
• acc23.dataset: Submodule that defines acc23.dataset.ACCDataModule, which is a pytorch lightning datamodule that takes care of importing and preprocessing the challenge data.
• acc23.postprocessing: Contains everything pertaining to postprocessing, i.e. going from raw model outputs to clean prediction CSV files. Important methods are acc23.postprocessing.eval_on_test_dataset and acc23.postprocessing.eval_on_train_dataset.
• acc23.explain: Everything related to explainability of the model's predictions. Important members are acc23.explain.VitExplainer which produces attention maps for vision transformers, and acc23.explain.shap, which approximates SHAP values in a model-agnostic way.
• acc23.utils: acc23.utils.train_model and other miscellaneous stuff.

The following modules also exist but the user shouldn't need to use them directly

• acc23.preprocessing: Contains everything pertaining to preprocessing. Used by acc23.dataset.ACCDataModule.
• acc23.constants: Constants about the dataset, e.g. the number of features or the name of the target columns.
• acc23.mlsmote: Implementation of the MLSMOTE dataset augmentation algorithm. Part of the preprocessing pipeline.

Submitting via acc23 CLI:

. ./secret.env && python3 -m acc23 submit -t "$TOKEN" out.csv dummy.ipynb

Troubleshooting

2023-04-26 Image corruption

data/images/CY60527_4_190006236104_2022_12_22_12_15_22.bmp is corrupted? PIL raises an OSError when loading... data/images/CY60527_4_190006236104_2022_12_22_12_11_20.bmp appears similar, so i just

cp data/images/CY60527_4_190006236104_2022_12_22_12_15_22.bmp data/images/CY60527_4_190006236104_2022_12_22_12_11_20.bmp

and called it a day

Contributing

Dependencies

• python3.10 or newer;
• requirements.txt for runtime dependencies;
• requirements.dev.txt for development dependencies.
• make (optional);

Simply run

virtualenv venv -p python3.10
. ./venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
pip install -r requirements.dev.txt

Documentation

Simply run

make docs

This will generate the HTML doc of the project, and the index file should be at docs/index.html. To have it directly in your browser, run

make docs-browser

Code quality

Don't forget to run

make

to format the code following black, typecheck it using mypy, and check it against coding standards using pylint.