Perceptron learning rule (Geron Ch 10, morie.fn hmpcpt)

w <- w + eta (y - y_hat) x with a step activation, run row by row in index order for at most n_iter epochs, stopping the moment an epoch makes no mistake. Converges iff the classes are linearly separable, which is why XOR never converges.

Usage

morie_geron_perceptron(X, y, eta = 1, n_iter = 10)

Arguments

X

Design matrix.

y

Labels in {0, 1}.

eta

Positive rate.

n_iter

Maximum epochs.

Value

List with w, weights, bias, predictions, accuracy, mistakes_per_epoch, converged, estimate, n.

Examples

morie_geron_perceptron(matrix(c(2, 1, 3, 2, -1, -2, -2, -1),
    nrow = 4, byrow = TRUE), c(1, 1, 0, 0), eta = 0.5, n_iter = 20)
#> $w
#> [1] 0.5 1.0
#> 
#> $weights
#> [1] 0.5 1.0
#> 
#> $bias
#> [1] -0.5
#> 
#> $predictions
#> [1] 1 1 0 0
#> 
#> $accuracy
#> [1] 1
#> 
#> $mistakes_per_epoch
#> [1] 1 0
#> 
#> $converged
#> [1] TRUE
#> 
#> $estimate
#> [1] 0.5 1.0
#> 
#> $n
#> [1] 4
#> 
#> $method
#> [1] "Perceptron rule w <- w + eta (y - y_hat) x with a step activation"
#>