So in my example, Ac is the name of a vector parameter… there’s Ac,Ac,Ac (there’s an even longer parameter Arbf which has 20 odd dimensions).
these are separate variables in the chain. When I do get(chain,:Ac) I get:
(Ac = ([0.32228709765451763 -0.3877980743116929; 0.3163860790999562 -0.38479165892268785; … ; 0.31895958580486605 -0.3156285003214462; 0.30888825934550446 -0.3198781895652254], [2.322569930200603 2.872963818099482; 2.314061819602352 2.8748381553809788; … ; 2.347355232722247 2.800025888639181; 2.340723325565215 2.7959559127543203], [-0.2897274052240421 0.36649297086057114; -0.2863375221649774 0.36842656904584226; … ; -0.25923055662025973 0.19406550561162855; -0.26198777592822214 0.20150952347443402]),)
So if you want to unparse that, this is a named tuple with one entry, Ac, the value is an unnamed tuple of three vectors…
if I do:
getindex.(get(chain,:Ac).Ac,200) I wind up with a 3 tuple:
(0.3032212499585117, 2.338802084753996, -0.25473893714523604)
I guess that might work fine for my purposes. Suppose I have a really high dimensional parameter though, like 1500, is there an advantage to having an Array rather than a high dimensional tuple?
in the end I want to call my predictor function with the values of the parameters:
where Ac and As and Arb are vectors because inside my predictor there’s linear algebra going on (dot products and rescalings and things).
Ultimately, it seems like inside a Chain is an array, and it would be great if I could just slice out the piece of the array that I am interested in, and look at that… so
chain.value[200,13:25,1] for example… except how do I figure out which are the dimensions associated with the stuff I want to slice out?