Deficiency one and concnetration robustness #964
Conversation
|
This needs to be updated to master. |
src/network_analysis.jl
Outdated
|
|
||
| for (c_s, c_p) in collect(Combinatorics.combinations(nonterminal_complexes, 2)) | ||
| # Check the difference of all the combinations of complexes. The support is the set of indices that are non-zero | ||
| supp = findall(!=(0), Z[:, c_s] - Z[:, c_p]) |
There was a problem hiding this comment.
The column difference could be stored in a pre-calculated vector from outside the loop like
@. tmp = @views Z[:, c_2] - Z[:, c_p]There was a problem hiding this comment.
More generally, seems like this could be written as a non-allocating loop that bails if a second match is found instead of continuing the search, and if a second match isn't found saves it in robust_species.
There was a problem hiding this comment.
Re the first comment: would I accomplish the same thing by adding @views in the supp line, so it becomes
supp = findall(!=(0), @views Z[:, c_s] - Z[:, c_p])Seems like allocating it entirely outside might take a lot of space (would have to be a 3D array, where each difference vector is indexed by (c_s, c_p))
For the second comment, what do you mean by second match? I don't immediately see a way to skip a check for a species we already know is robust because we can't know that it's in the support of a difference of two complexes until we calculate it
|
@vyudu LGTM. Feel free to merge at your leisure. |
Concentration robustness for deficiency one networks. Relies on strong and terminal linkage classes.