Elsewhere, I've been using an ODE solver that returns an answer that contains the erfi(z)
function, where z might be complex.

This function can return quite large numbers even for moderately large values of z, do you know of a
way to avoid, or manage, this ?

Yes, however what type of A.I; a rather large and power hungry variety or a nimble edge computing device. Once you've trained a Neural Network , for instance, then you're able to use the known weights in a very power efficient way. Even if a certain set of weights doesn't meet all of your requiremen...

What should the FreeBasic community use ChatGPT, or similar, for ? There's some low lying fruit; like developing a cross platform GUI designer, with a myriad of features, including embedded animated graphics. Translating libraries from other languages. Hopefully we don't need a nuclear fission react...

A little graphics plotting program I've been writing, with the help [65%] of ChatGPT and hindrance [30%]; my contribution is indeterminate. To get specular reflection from the surface of the [sinc] function, the calculation of normal's is most likely required. ' ' ' Freebasic : luxan ' sciwiseg@gmai...

Star Trek "The search for Spock"

Scotty:
The more they overthink the plumbing, the easier it is to stop up the drain.


Kruge:
No tricks, Kirk. You have one minute.

Luxan
You have ....

So, we've explored your development of the switchnet 4 , from the initial concept through to code that produces sensible results. If I go back to the original illustration of switchnet 4, where the standard fast WHT is shown, and I use that in the code will I retain the functionality of your evolved...

Single-threaded parallelisation, using Freebasic , sounds ambitious .

Thank you for informing me. If possible, I keep to the readily available functions and operators of the underlying language, it's generally easier to determine what's going on with the code that way. Also FreeBasic has quite good documentation. During the development phase I Iterate through many cod...

Your Sparse Net certainly incorporates a few extras, like the anti aliased line drawing.

At present my mind is probably over fitted to previous knowledge and pruning new
knowledge; a long learning curve is predicted.

At least until such time as there are fewer out standing tasks to complete.

It's option 0: 0 .

SparseNet, is this new to the Freebasic forum ?

It's just your option 1. 0, leaves me puzzled; do you mean to dynamically
prune the NN connections .

I've noticed my time stamp, and yours, we appear to be on opposite sides
of the world.

You're being whimsical , right ? There are a few things I'm attempting to do , one in particular though tends to preempt the others; home renovation . Eventually I'll bring all of the bits and pieces of Switch nets together into, hopefully, a functioning whole that I , and anyone else, can comprehend.

I updated some code I'd been using previously to examine the WHT. With L=4, N=16, hs_pow2_shl = 2, it's apparent that the first two passes are skipped and the remaining are evaluated. Is setting hs_pow2_shl = 3, ever considered ? Elsewhere I examined this via your SwNet4.recall(), with certain piece...

For the last, illustrated, example pow2_shl = 1, within your SwNet4.recall it's const pow2_shl = 2 .
Hence N>8 is required before the last pass[es] of the wht will be evaluated .

For the wht, where multiplications aren't involved or twiddle factors, your shorter sub using while statements is a useful approach. However being able to compare this to the more traditional representation helps illuminate what's being attempted. Now, so far, I've illustrated the situation for an 8...