loadmess

I recently noticed that jit.normalize method is based on addition and doesn't work with negative numbers, which doesn't work for what I need because any number that is zero gets added by something.

So I began making my own version using jit.op.

I'm working with float32 matrices, that contain positive and negative numbers, so the pseudocode would be something like:

3) Find the normalization ratio [1. / max. value] in order to normalize between (-1. and 1.)

There's one issue that I'm missing, the matrix needs to be sent twice to get the expected results.

Is there any one jitter object or method that would get the expected results?

comment

jit.cellblock

jit.normalize

jit.op

jit.max

jit.abs

inlet

outlet

message

jit.matrix

Hello Maxers,
I recently noticed that jit.normalize method is based on addition and doesn't work with negative numbers, which doesn't work for what I need because any number that is zero gets added by something.
So I began making my own version using jit.op.
I'm working with float32 matrices, that contain positive and negative numbers, so the pseudocode would be something like:
1) absolute the matrix
2) find maximum of (abs matrix)
3) Find the normalization ratio [1. / max. value] in order to normalize between (-1. and 1.)
4) Apply normalization: [Matrix * Ratio]

There's one issue that I'm missing, the matrix needs to be sent twice to get the expected results.
What I'm I missing?
Is there any one jitter object or method that would get the expected results?

Hello Maxers,
I recently noticed that jit.normalize method is based on addition and doesn't work with negative numbers, which doesn't work f

jit-normalize-method-adding-vs-multiplying

I found the solution. Some values need to be initialized every time.

Just replace the previous normalize abstraction with this one.

I found the solution. Some values need to be initialized every time.
Just replace the previous normalize abstraction with this one.


jit.normalize method: adding vs multiplying