WildfireIA

Official implementation for WildfireIA, a benchmark for predicting whether a wildfire will escape initial attack from public information available at fire discovery time.

Code is maintained by the Responsible AI Lab at Florida State University.

Dataset release:

https://huggingface.co/datasets/WildfireIA/Anonymous-WildfireIA

The dataset repository contains canonical benchmark tables and Croissant metadata. It does not contain model-ready caches. After cloning this code repository, place the canonical tables at the path below and regenerate caches with dataloader.py.

Quick Start

Clone this repository:

git clone https://github.com/LabRAI/WildfireIA.git
cd WildfireIA

Install Python packages:

python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

PyTorch GPU wheels depend on the local CUDA version. If the default torch installation is not compatible with your system, install PyTorch following the official PyTorch instructions, then rerun pip install -r requirements.txt.

Download the Hugging Face dataset into a temporary folder:

git lfs install
git clone https://huggingface.co/datasets/WildfireIA/Anonymous-WildfireIA hf_data

Copy the canonical tables into this repository:

mkdir -p data/canonical/raw_feature_tables
rsync -a hf_data/data/canonical/raw_feature_tables/ data/canonical/raw_feature_tables/

The expected path is:

data/canonical/raw_feature_tables/

Generate model-ready caches:

python dataloader.py \
  --base_dir . \
  --canonical_dir data/canonical/raw_feature_tables \
  --output_dir data/cache/model_ready \
  --task ia_failure \
  --representation all \
  --weather_days 5 \
  --input_protocol all \
  --overwrite

Run one baseline:

python train.py \
  --base_dir . \
  --task ia_failure \
  --experiment_type smoke \
  --representation tabular \
  --weather_days 5 \
  --input_protocol all \
  --model xgboost \
  --seed 553371 \
  --overwrite

The output is written to:

experiments/ia_failure/smoke/tabular/weather5_all/xgboost_seed553371/

Important output files include:

config.json
metrics.json
predictions_val.parquet
predictions_test.parquet

Main Experiment Pattern

Experiment outputs follow this directory format:

experiments/{task}/{experiment_type}/{representation}/weather{days}_{protocol}/{model}_seed{seed}/

Example full-input XGBoost run:

python train.py \
  --base_dir . \
  --task ia_failure \
  --experiment_type full \
  --representation tabular \
  --weather_days 5 \
  --input_protocol all \
  --model xgboost \
  --seed 553371 \
  --overwrite

Example spatial neural baseline:

python train.py \
  --base_dir . \
  --task ia_failure \
  --experiment_type full \
  --representation spatial \
  --weather_days 5 \
  --input_protocol all \
  --model swin_unet \
  --seed 553371 \
  --max_epochs 100 \
  --batch_size 64 \
  --early_stop_patience 15 \
  --sampling_strategy weighted \
  --standardize_channels \
  --overwrite

Scripts

• pipeline.py: optional raw-data canonicalization script. Users who download the canonical tables from Hugging Face do not need to run it.
• dataloader.py: converts canonical tables into model-ready caches.
• train.py: trains tabular, temporal, spatial, and spatiotemporal baselines.

Supported Models

• Tabular: logistic_regression, xgboost, mlp
• Temporal: gru, tcn, transformer
• Spatial: resnet18_unet, resnet50_unet, swin_unet, segformer
• Spatiotemporal: convlstm, convgru, predrnn_v2, utae, swinlstm, resnet3d

License

This code is released under the MIT License.