should reduce time for feature generation

limit the number of operations done

Works on initial test cases

It works locally, but not on the cluster

I was too agressive with removing features, fixed that

made an explicit iterate function to do this recursively

More feature elimination steps. Clear out register when doing generation/proejction

Forgot to set_value in new sis checsk

Division/multiplcation nodes cant have division features/inverse

empyt unit check

Check if this helps with memory problems

ii= 0 -> int ii = 0

Do not use all_gather, but sends/recv. see if that fixes the bug when SIS is too large

a typo and test_value_ptr()/set_test_value() only redirects if _selected is false for feature node

This should hopefully reduce the codes memory footprint

Forgot some comment lines

Don't make vectors of vectors, but transfer directly

Each process picked its own train/test set, which is bad

typos when introducing the test set led to problems

Did them one after another, adding both here now

SIS updated to use scores to check similarity and only send phi's that are needed

For the distribution minor bug led to hangs

it should not always be true

all lower rungs stored on processes, this was causing issues with set_value if not done this way

Node and OperatorNode get virtual destructors

More complincated tests are needed

phi stored across all nodes not stored on all nodes

Possible to resize based off of rung

Initial commit towards resizing temporary storage arrays

Checks for cancelations and duplicate features

Removed _value_ptr and _feat_value_ptrs from OperatorNode
Allow set_value() and get_value_ptr() to accept offest and work directly on offset
value_ptr only thing to call set_value

Added type condition tests (initial simplification)
Added som more documenatation