CURE(Context management via Uncertainty-aware admission and Redundancy-aware Eviction)

Install

pip install -r requirements.txt

Run Main 7-Dataset Setting

bash scripts/run_main_7datasets.sh

The main per-dataset tau values are stored in:

config/expected_main_taus.json

Run One Dataset

python scripts/run_tabicl_cure.py \
  --dataset noaa \
  --tau 0.4 \
  --data-root "$PWD/data/uspds" \
  --output-root "$PWD/results" \
  --device cuda \
  --threads 4

Output

Results are saved under:

results/{dataset}/

Each dataset directory contains:

summary.json
run_config.json
step_metrics.csv

Directory Structure

.
├── data/
│   └── uspds/
├── config/
│   └── expected_main_taus.json
├── scripts/
│   ├── run_tabicl_cure.py
│   └── run_main_7datasets.sh
├── src/
│   └── stream/
├── requirements.txt
└── results/

Checkpoint

The TabICLv2 model checkpoint is not included in this bundle. Running the code requires either:

• online checkpoint resolution/download through the tabicl package, or
• a local checkpoint path supplied through TABICL_MODEL_PATH

For offline use, set:

export TABICL_MODEL_PATH=/path/to/tabicl-classifier-v2-20260212.ckpt