Skip to contents

Fold model dimensions to reduce data size

Source: R/fold.R
fold_model.Rd

Reduces redundant dimensions in parameters where data is constant across those dimensions. Only folds high-cardinality indexing dimensions (timeslice, optionally region/year) while preserving structural dimensions (tech, comm, etc.).

Usage

fold_model(
  model,
  fold_spec,
  tolerance = 1e-10,
  verbose = TRUE,
  .progress = "auto"
)

Arguments

model

A multimod model object

fold_spec

Data frame specifying fold configuration. Must have columns: param_name, dim_to_fold, validation_mapping. Use create_fold_spec() to generate proper fold specification with entity-specific mappings.

tolerance

Numeric tolerance for considering values equal (default: 1e-10)

verbose

Logical; print progress messages (default: TRUE)

.progress

Progress bar control: "auto" (default) shows a progressr bar for large models, TRUE forces it on (requires progressr), FALSE disables it.

Value

Modified model with folded parameters

Details

The algorithm uses per-group validation: for each combination of the other dimensions, it checks whether the values are identical. The test is all or nothing: if any group shows variation, the dimension is not folded.

Folding analyzes each parameter to identify dimensions that do not provide variation in the data. For each parameter, it groups by all other dimensions and checks whether the values are constant within each group (within tolerance).

Example: For pStorageInpEff[stg, comm, region, year, timeslice] testing timeslice:

  • Groups by (stg, comm, region, year)

  • For each group, checks whether all timeslice values are identical

  • Folds only if every group has constant values

Sparse data is handled: a parameter with values only for peak hours (one timeslice out of 17) is tested on the values that exist, which must be constant within each group.

The fold creates three slots per parameter:

  • data: Original complete data (preserved)

  • folded_data: Collapsed data with reduced dimensions

  • active_dims: Dimensions remaining after fold

Safeguards:

  • Only the dimensions named in fold_spec are folded

  • Dimensions that identify an entity (for an energyRt model: tech, comm, sup, dem, stg, trade) are never folded

  • Skips parameters where all dimensions would be folded (would become scalar)

  • Validates consistency before folding

  • Updates equation ASTs to match reduced dimensions

Examples

m <- example_models$energyRt$multimod
fold_spec <- create_fold_spec(m, fold_dims = "timeslice", verbose = FALSE)
# Review and edit fold_spec to ensure correct validation mappings
m_folded <- fold_model(m, fold_spec = fold_spec, verbose = FALSE)
get_fold_summary(m_folded, format = "text")