Freely defining aggregation over nodes

[guyko81]

It would be interesting to allow multiplication effect from one node to another. Example: in time series analysis if the Sundays are always closed, so the sales is always 0, it's not possible to create a clean 0 prediction from pure additive rules. If a node could affect the final result through a multiplier logic, it would be possible to learn closed store cases.

set_aggregate_effect(0, [1,1], 'x_3*(x_0+x_1+x_2)') where the first argument is the input layer, second argument is the target layer and node, third argument is the equation for the aggregation of the input nodes.

[1991jhf]

you can do log transform to make it multiplicative.

[guyko81]

You're right - stupid me. Although the solution for any function will be very convoluted. Let me put something together before closing this issue.

[guyko81]

I hope this correct, but even if I made a mistake everyone gets the point. Doable, but not easy :)

$$ x_4(x_1+x_2+x_3) $$

model = KAN(width=[4,2,1,1], grid=6, k=3) model.train(dataset, opt="Adam", lr=0.01, steps=1)

x1+x2+x3

model.remove_edge(0,0,1) model.remove_edge(0,1,1) model.remove_edge(0,2,1)

keeping x4

model.remove_edge(0,3,0)

forcing linear, so it surely simple sum

model.fix_symbolic(0,0,0,'x') model.fix_symbolic(0,1,0,'x') model.fix_symbolic(0,2,0,'x') model.fix_symbolic(0,3,1,'x')

adding logarithm

model.fix_symbolic(1,0,0,'log') model.fix_symbolic(1,1,0,'log')

exponential

model.fix_symbolic(2,0,0,'exp')

$$ \exp(\log(x_4) + \log(x_1 + x_2 + x_3)) $$