Downloads · 30 days
0
experiment-5-v1-autogluon-bias-corrector-model
cs-labs/experiment-5-v1-autogluon-bias-corrector-model
experiment-5-v1-autogluon-bias-corrector-model is a machine learning model from cs-labs. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
AutoGluon 1.5 Extreme weighted ensemble (4-level stacking) for daily high temperature bias correction at RKSI.
Access
Public
Updated Mar 31, 2026
Repo size
21.8 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pkl13.6 GB · 62%
At a glance
- Access
- Public
- Created
- Mar 31, 2026
- Updated
- Mar 31, 2026
- SHA
- aa6c9b22
Model card
From the Hugging Face model README
Experiment 5 v1 - AutoGluon Bias Corrector
Model
AutoGluon 1.5 Extreme weighted ensemble (4-level stacking) for daily high temperature bias correction at RKSI.
Station
- Target: RKSI (Incheon International Airport, South Korea)
- Coordinates: 37.4692N, 126.4505E
Results (Holdout Year 2025)
- MAE: 0.5013 C
- Bias: -0.0058 C
- RMSE: 0.6707 C
- Integer Accuracy: 60.0%
- Raw DiT MAE (before correction): 2.91 C
- Improvement: 5.8x error reduction
Best Model: WeightedEnsemble_L4_FULL
Top Models (Leaderboard)
| Model | Val Score (neg MAE) | Stack Level |
|---|---|---|
| WeightedEnsemble_L4 | -0.8488 | 4 |
| WeightedEnsemble_L3 | -0.8505 | 3 |
| RealTabPFN-v2_c1_BAG_L3 | -0.8560 | 3 |
| RealTabPFN-v2_r11_BAG_L3 | -0.8572 | 3 |
| RealTabPFN-v2_c1_BAG_L2 | -0.8581 | 2 |
Training Configuration
- Preset: extreme
- Time limit: 14400s (4h)
- Bag folds: 8
- Stack levels: 3
- Refit full: True
- Training data: 3652 rows (2015-2024)
- Validation data: 365 rows (2025)
Features
DiT model predictions (T2m max, step-wise temps, pressure, winds, humidity, radiation), Weather Underground hourly observations (temperature, dewpoint, humidity, pressure, wind, visibility, cloud cover), and derived temporal features (warming rates, overnight drop, dew depression, etc.)
Hardware
- 192 CPUs, 8x NVIDIA B200 GPUs
- Training time: ~42 min
- Created
- Mar 31, 2026
- Updated
- Mar 31, 2026