Walks each equation's AST set-at-a-time and returns the LP matrix entries together with the right-hand side implied by the equation's constant terms.
Usage
build_triplets(
model,
col_index = NULL,
row_index = NULL,
equations = NULL,
on_duplicate = c("error", "last", "first"),
verbose = FALSE,
chunk_rows = 1000000L
)Arguments
- model
A multimod model with data attached.
- col_index, row_index
Index tables from
build_col_index()/build_row_index(). Built on demand when not supplied.- equations
Optional character vector limiting which equations to process; useful when bringing the backend up equation by equation.
- on_duplicate
What to do when a parameter stores the same index tuple with more than one distinct value.
"error"(default) refuses, because the lookup is genuinely ambiguous;"last"takes the last, matching what GAMS and Pyomo do silently;"first"takes the first. Either non-default choice reports the parameter and the number of tuples affected.- verbose
Logical; report per-equation term counts and timings.
- chunk_rows
Integer; evaluate each equation over at most this many LP rows at a time. Each LP row's terms are independent, so slicing by row is exact; what it bounds is the pre-aggregation intermediate of a
sum()(outer rows x inner tuples), which on a large full-year model otherwise allocates tens of gigabytes in one piece. The aggregated per-chunk triplets concatenate without renumbering (.kcarries the global row id).