Runs structural checks on a knowledge graph before it is used to build a
model. The flags use the same three levels as the
assessment pipeline: block (the graph cannot be used as
it is), review (a psychometrician should look) and info.
Arguments
- kg
a knowledge graph from
mv_kg_build()oras_mv_kg().- min_items
smallest number of items per concept before a
few_itemsflag.
Value
A tibble of class mv_kg_check with columns check, level,
nodes (the concepts or items involved) and message; zero rows when
nothing was found.
Details
The checks:
cycle(block): Prerequisite edges form a cycle, so no learning order exists.self_loop(block): A concept is listed as its own prerequisite.few_items(review): A concept is measured by fewer thanmin_itemsitems, too few to estimate it as its own dimension.unmeasured(review): A concept is measured by no item. This is fine for a concept that only organizes the graph, but it cannot be scored.same_items(review): Two concepts are measured by exactly the same items, so the data cannot tell them apart.redundant_prerequisite(info): A prerequisite edge is implied by a longer path (A before B before C makes A before C redundant).isolated(info): A concept has no relations to other concepts, in a graph where other concepts do have relations.duplicate_edge(info): The same edge is listed more than once.
Examples
mv_kg_check(fractions_kg)
#> <mv_kg_check> 0 block, 0 review, 1 info
#> * [info] isolated: compare has no relations to other concepts.
mv_kg_check(fraction_subtraction_kg)
#> <mv_kg_check> 0 block, 1 review, 0 info
#> * [review] few_items: column_borrow is measured by 2 item(s); at least
#> 3 are needed to estimate it as its own dimension.