How ipl-uv/climx works
ClimX is a climate-model emulation benchmark: users prepare forcing and target data, train an emulator, predict daily climate fields, derive climate indices, and submit those indices for regional extreme-aware scoring. The README documents an end-to-end notebook, baseline training, Hugging Face data, Kaggle formatting, and an optional probabilistic track. Source excerpts confirm the training, preprocessing, model, index, and submission components; unsampled integrations are shown only where documented, and the optional uncertainty track has no implementation path established by the tree.
Main entry points: train.py Baseline training · emulator.py Emulator interface · preprocessing.py Climate preprocessing · climatology_model.py Climatology baseline
Components 19
Benchmark workflow
- Baseline training
train.py - Dataset access
src/utils/hugging_face_utils.py - Climate preprocessing
src/data_preprocessing/preprocessing.py - Transform utilities
src/data_preprocessing/preprocessing_utils.py
Emulator baselines
- Emulator interface
src/models/emulator.py - Climatology baseline
src/models/climatology_model.py - Neural baseline
src/models/nn_model.py - Graph baseline
src/models/gnn_model.py - LPS baseline
src/models/lps_model.py
Indices and scoring
- Climate indices
src/indices_xclim.py - Index metadata
src/index_metadata.py - Evaluation metrics
src/metrics.py - Regional nNSE
rnNSE.py
Submission and inspection
- Kaggle formatting
src/utils/kaggle_submission.py - Climate visualization
src/visualization/visualize_data.py - Probabilistic track
Other components
- Benchmark participant
- Hugging Face dataset
- Kaggle tracks
Connections 21
- Benchmark participant → Baseline training: trains baselines
- Hugging Face dataset → Dataset access: provides data
- Baseline training → Dataset access: opens dataset
- Baseline training → Climate preprocessing: prepares data
- Climate preprocessing → Transform utilities: applies transforms
- Baseline training → Climatology baseline: selects baseline
- Baseline training → Neural baseline: selects baseline
- Baseline training → Graph baseline: selects baseline
- Baseline training → LPS baseline: selects baseline
- Climatology baseline → Emulator interface: implements
- Neural baseline → Emulator interface: implements
- Graph baseline → Emulator interface: implements
- LPS baseline → Emulator interface: implements
- Benchmark participant → Climate indices: derives indices
- Climate indices → Index metadata: uses definitions
- Benchmark participant → Kaggle formatting: formats predictions
- Kaggle formatting → Kaggle tracks: prepares submission
- Kaggle tracks → Regional nNSE: scores main track
- Benchmark participant → Climate visualization: inspects climate data
- Kaggle tracks → Probabilistic track: offers UQ track
- Probabilistic track → Kaggle tracks: evaluates CRPS