Index and other optimisations #922
Commits on Feb 23, 2024
-
Indices: optimise _calc_endemism_absolute
Take advantage of the label hash global precalc, and use hash aliases instead of refs.
-
-
TreeNode.pm: use a linear scan for get_hash_lists_below
Might as well avoid any recursion overheads.
-
Add an array args version of set_basedata_ref
And stop throwing errors when ref is undefined in get_basedata_ref.
-
Trees: clone_without_caches also clears parameters
This was we avoid cloning basedata refs, analysis args and the like.
-
Common::get_zscore_from_comp_results - avoid a lot of grepping
No need to find the index names when they are in the base_list_ref already. Also use refaliasing to avoid some derefs and declutter loop variables.
-
optimise Tree::convert_comparisons_to_significances
Passing in the base list allows fewer grep comparisons. This makes a large difference when there are many lists with many keys.
-
optimise Spatial::convert_comparisons_to_significances
Passing in the base list allows fewer grep comparisons. This makes a large difference when there are many lists with many keys.
-
Indices: add a hierarchical mode flag
This allows future optimisations when calculating indices for cluster trees.
-
Indices: Support hierarchical calculations
This allows several indices to be optimised when calculated for cluster nodes, providing they are done starting from the tips. PE has been optimised in this commit.
-
Indices: calc_labels_not_on_tree: return early if nothing to work with
Avoids a lot of hash creation and deletion with large datasets.
-
Indices: add a hierarchical variant of get_path_lengths_to_root_node
Speeds up PD calcs for cluster trees.
-
Indices: _calc_endemism_hier_part: avoid some method calls
Use a treenode method that caches, rather than repeatedly calling methods to get the same answer.
-
-
Commits on Feb 24, 2024
-
Indices: refactor hierarchical node details
It is cleaner to pack the node and child names in their own structure. That also enables later additions without adding yet more top level arguments.
-
TreeNode::add_to_lists: optimise
Use direct assignment if starting with empty list.
-
compare_lists_by_item: lift a var outside the loop
This can be a _very_ hot loop so even small differences add up.
-
Indices: cache the current results from each sub
These are cleared as we go to avoid leakage.
-
Indices: reuse whole and central endemism results when appropriate
If the second neighbour set is empty then the whole and central variants return the same results. So short circuit in these cases.
-
-