feat: Graph-EFM ensemble forecasting model - #712

Draft

Sir-Sloth-The-Lazy wants to merge 56 commits into

Sir-Sloth-The-Lazy:feat/graph-efm-model

Commits on Jun 21, 2026

feat: add latent encoder/decoder infrastructure for Graph-EFM port
Sir-Sloth-The-Lazy
committed
Update neural_lam/models/latent/constant_encoder.py

Sir-Sloth-The-Lazy
and
joeloskarsson
committed
feat: address review feedback on latent encoder/decoder infra
Sir-Sloth-The-Lazy
committed
style: apply black formatting to hierarchical latent modules
Sir-Sloth-The-Lazy
committed
feat: add GraphEFM single-step probabilistic predictor (PR-6)
Sir-Sloth-The-Lazy
committed
refactor: hard-code constrained latent GNN types, reuse existing flags for the rest
Sir-Sloth-The-Lazy
committed
fix: satisfy interrogate docstring coverage from upstream main
Sir-Sloth-The-Lazy
committed
docs: describe current code only in docstrings and comments
Sir-Sloth-The-Lazy
committed
docs: convert latent module and make_gnn_seq docstrings to numpy style
Sir-Sloth-The-Lazy
committed
refactor: skip on-mesh/intra-level processing explicitly instead of IdentityModule
Sir-Sloth-The-Lazy
committed
docs: align ConstantLatentEncoder.compute_dist_params docstring with base class
Sir-Sloth-The-Lazy
committed
Update neural_lam/models/latent/base_decoder.py

Sir-Sloth-The-Lazy
and
joeloskarsson
committed
docs: expand parameter descriptions in latent decoder docstrings
Sir-Sloth-The-Lazy
committed
docs: clarify role of grid update MLP in latent decoder forward
Sir-Sloth-The-Lazy
committed
docs: explain mean/std chunking of decoder output parameters
Sir-Sloth-The-Lazy
committed
refactor: split Graph-EFM into hierarchical and flat subclasses
Sir-Sloth-The-Lazy
committed
refactor: avoid processor terminology in Graph-EFM parameters
Sir-Sloth-The-Lazy
committed
docs: document all constructor parameters in Graph-EFM subclasses
Sir-Sloth-The-Lazy
committed
refactor: remove sample_obs_noise option from Graph-EFM
Sir-Sloth-The-Lazy
committed
refactor: extract shared graph-setup helpers to remove duplication
Sir-Sloth-The-Lazy
committed
refactor: give descriptive names to Graph-EFM grid embedding methods
Sir-Sloth-The-Lazy
committed
refactor: drop redundant last_state alias in Graph-EFM forward
Sir-Sloth-The-Lazy
committed
Update neural_lam/utils.py

Sir-Sloth-The-Lazy
and
joeloskarsson
committed

Commits on Jun 24, 2026

Commits on Jun 27, 2026

Merge branch 'main' into feat/latent-encoder-decoder-infra
Sir-Sloth-The-Lazy
authored

Commits on Jul 5, 2026

feat: move training loss onto Forecaster, add probabilistic interface
Sir-Sloth-The-Lazy
committed

Commits on Jul 6, 2026

Address PR review: clarify scoring-rule wording, rename score_fn to score_metric
Sir-Sloth-The-Lazy
committed

Commits on Jul 8, 2026

Commits on Jul 10, 2026

Commits on Jul 11, 2026

Commits on Jul 14, 2026

refactor: strip likelihood/ELBO assembly out of GraphEFMPredictor
Sir-Sloth-The-Lazy
committed

Commits on Jul 15, 2026

Merge branch 'main' into feat/latent-encoder-decoder-infra
Sir-Sloth-The-Lazy
authored

Commits on Jul 20, 2026

Commits on Jul 21, 2026

Address PR review: remove now-unnecessary TYPE_CHECKING guard on BaseDatastore
Sir-Sloth-The-Lazy
committed

Commits on Jul 25, 2026