solve_system: Solve the linear system between a hidden layer and the...
In mugiro/elm: Extreme Learning Machine for regression and classification

Description Usage Arguments Details Value

Solve the linear system h [n x l] * w_out [l x c] = y [n x c]

1	solve_system(object, ...)

`h`	a `matrix` of dimensions [n x l], with the outputs of the hidden layer
`solve`	a `logical` to set if w_out has to computed. If FALSE, only the two correlation matrices are returned
`y`	a `matrix` of dimensions [n x c] or a `vector` if c = 1, with the output values

The Moore-Penrose generalized inverse is found by using orthogonal projection. The two correlation matrices hh = t(h) * h and ht = t(h) * y are computed. The final squared linear system solved is: hh [l x l] * w_out [l x c] = ht [n x c] Orthogonal projection is used to compute the generalized inverse of the matrixUse orthogonal projection - correlation matrices

w_out a matrix of dimensions [l x c] with the output weights

mugiro/elm documentation built on May 23, 2019, 8:21 a.m.