ParamsArray type #219
ParamsArray type #219
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## devel #219 +/- ##
==========================================
+ Coverage 71.69% 74.14% +2.44%
==========================================
Files 52 52
Lines 2120 2197 +77
==========================================
+ Hits 1520 1629 +109
+ Misses 600 568 -32 ☔ View full report in Codecov by Sentry. 🚨 Try these New Features:
|
| a.param_ptr == b.param_ptr && | ||
| a.linear_indices == b.linear_indices | ||
|
|
||
| Base.hash(a::ParamsArray, h::UInt) = hash( |
There was a problem hiding this comment.
What do we need that for?
There was a problem hiding this comment.
I added == and hash() so that ParamsArray could be efficiently used as a key in AbstractDict (particularly for matrices large dimensions).
The reason why I need it -- when I construct my regularized multi-group SEM, some terms might be duplicate and could therefore be merged, but it's hard to detect in advance (e.g. I may have series of replicate experiments that have either 100% identical design or each of the series may have some unique samples not present in the other series). So with Dict{ParamsArray, SEMTerm}() I can automatically detect duplicate terms.
SEM.jl does not need/use it, so in principle I can move it to my code. But that's relatively little code, and by keeping it in SEM.jl we avoid "type piracy".
There was a problem hiding this comment.
Alright, seems reasonable - let's keep it here then.
|
Thanks a lot, great PR - I just had some small comments, otherwise, i would merge it. |
no declarations, so import is not required
replaces RAMMatrices indices and constants vectors with dedicated class that incapsulate this logic, resulting in overall cleaner interface A_ind, S_ind, M_ind become ParamsArray F_ind becomes SparseMatrixCSC parameters.jl is not longer required and is removed
Thank you for the review! I've addressed your comments. |
Maximilian-Stefan-Ernst
merged commit 5b9327e
into
StructuralEquationModels:devel
This PR is another subset of #193 changes and adds
ParamsArraystype to simplify RAM specification.ParamsArray{T,N}object defines N-dimensional array, whose elements are either constants of typeTor parameters (specified as indices in parameters array).materialize(pararr, parvals),materialize!(arr, parrarr, parvals)andsparse_materialize(pararr, parvals)are used to create/update existing arrays with theparvalsparameter values according to thepararr::ParamsArrayspecification.It also supports fast sparse gradients by the parameters vector via
sparse_gradient!().For lower level iteration over parameter references in an array, one can use the
param_occurences(pararr, i)method that returns the linear indices of i-th param occurences.The PR switches
RAMMatricesto use theParamsArrayfor A and S matrices and M vector.