algorithmic modeling for Rhino
strokes of a drawing to vector field
tangent vectors on strokes learnt by resilient propagation algorithm to construct a vector field function from it;
learning time 10 minutes (8 cores)
Tags:
Comment
Hi!
Can you explain the workflow a bit more? you mentioned it's gonna be part of Octopus?
Best regards
Shahrokh
The 'field' is a collection of Bipolar Sigmoid Functions. Their coefficients get altered by the algorithm.
The network has
2 input nodes = x|y of point (a point on the curves)
2 output nodes = x|y of assigned direction (the tangent)
and a number of hidden nodes which are fully connected.
Dividing the curves then gives you a number of samples (pairs of inputs with known desired pairs of output values) to figure out some coefficients which minimize the error between inputs and outputs.
The rest is simple RungeKutta4 integration with the trained 'field' giving directions.
A big drawback is that it's not PI-invariant. I cant figure an efficient way to make it ignore the orientation of the input vectors.
Compared to just interpolating the vectors it's also dead slow and inaccurate. But it's an intuitive way to get a grip on what those networks can and cannot do well.
How is the field defined? Is it a sampling grid? A collection of polynomials? A collection of trig functions?
beautiful
schiele's standing nude with stockings btw
Oh now I understand. Your explanation flew right by me but makes sense now. Trippy. Interesting.
© 2019 Created by Scott Davidson. Powered by
You need to be a member of Grasshopper to add comments!
Join Grasshopper