gradient-descent ( X y params lr n-iters -- history params )

Vocabulary

tensors.demos.private

Inputs

X	an object
y	an object
params	an object
lr	an object
n-iters	an object

Outputs

history	an object
params	an object

Definition

USING: accessors kernel math sequences tensors ;

IN: tensors.demos.private

:: gradient-descent
    ( X y params lr n-iters -- history params )
    lr y shape>> first / :> batch-lr
    { n-iters } zeros :> history X transpose :> X-T params
    n-iters [
        swap dup :> old-params batch-lr X-T X old-params matmul
        y t- matmul t* t-
        :> new-params X y new-params compute-cost
        swap history set-nth new-params
    ] each-integer history swap ;