Shade by avara submission

[kevin-romer]

Summary

• Add benchmark_exp/Run_SHADE_by_Avara.py — a new pretrained anomaly detector submission based on SHADE (Sparse Hybrid Anomaly Detection Engine), a hosted time-series anomaly detection endpoint
• Update uni_mergedTable_VUS-PR.csv and multi_mergedTable_VUS-PR.csv with SHADE evaluation results
• SHADE achieves new state-of-the-art VUS-PR results on both TSB-AD-U and TSB-AD-M

Results

TSB-AD-U (Univariate)

Rank	Model	VUS-PR	Pretrained
1	SHADE	0.59	Yes
2	TSPulse (FT)	0.55	Yes
3	Time-RCD	0.52	Yes

TSB-AD-M (Multivariate)

Rank	Model	VUS-PR	Pretrained
1	SHADE	0.50	Yes
2	xLSTMAD	0.39	No
2	TSPulse (FT)	0.39	Yes

Running Instructions

pip install TSB-AD

Set endpoint credentials:

export SHADE_API_KEY='<provided privately>'
export SHADE_BASE_URL='https://shade.avara-ai.com'

Run the detector through the TSB-AD benchmark runner:

python benchmark_exp/Run_SHADE_by_Avara.py \
  --file_lsit Datasets/File_List/TSB-AD-U-Eva.csv \
  --data_direc /path/to/TSB-AD-U \
  --AD_Name SHADE_Lambda_U \
  --score_dir eval/score/uni \
  --save_dir eval/metrics/uni \
  --save true \
  --overwrite true

For multivariate evaluation:

python benchmark_exp/Run_SHADE_by_Avara.py \
  --file_lsit Datasets/File_List/TSB-AD-M-Eva.csv \
  --data_direc /path/to/TSB-AD-M \
  --AD_Name SHADE_Lambda_M \
  --score_dir eval/score/multi \
  --save_dir eval/metrics/multi \
  --save true \
  --overwrite true

Note: Run_SHADE_by_Avara.py connects to a hosted SHADE endpoint. Please reach out privately for the SHADE_API_KEY required for evaluation. The script sends only the input values, sanitized series name, split name, and train-prefix length to the endpoint. Labels are not sent to the endpoint and are used only locally by the TSB-AD evaluation code.

Files Changed

• benchmark_exp/Run_SHADE_by_Avara.py
• uni_mergedTable_VUS-PR.csv
• multi_mergedTable_VUS-PR.csv