args <- commandArgs(TRUE) if (length(args) > 0L) { method <- args[1] } else { method <- "simple" } if (length((args) > 1L)) { momentum <- as.double(args[2]) } else { momentum <- 0.0 } epochs <- 50L attempts <- 25L # library(CVEpureR) # path <- paste0('~/Projects/CVE/tmp/logger_', method, '.R.pdf') library(CVE) path <- paste0('~/Projects/CVE/tmp/logger_', method, '.C.pdf') # Define logger for `cve()` method. logger <- function(epoch, attempt, L, V, tau) { # Note the `<<-` assignement! loss.history[epoch + 1, attempt] <<- mean(L) if (epoch == 0) { error <- NA } else { error <- norm(V %*% t(V) - V_last %*% t(V_last), type = 'F') } V_last <<- V error.history[epoch + 1, attempt] <<- error tau.history[epoch + 1, attempt] <<- tau # Compute true error by comparing to the true `B` B.est <- null(V) # Function provided by CVE P.est <- B.est %*% solve(t(B.est) %*% B.est) %*% t(B.est) true.error <- norm(P - P.est, 'F') / sqrt(2 * k) true.error.history[epoch + 1, attempt] <<- true.error } pdf(path) par(mfrow = c(2, 2)) for (name in paste0("M", seq(5))) { # Seed random number generator set.seed(42) # Create a dataset ds <- dataset(name) X <- ds$X Y <- ds$Y B <- ds$B # the true `B` k <- ncol(ds$B) # True projection matrix. P <- B %*% solve(t(B) %*% B) %*% t(B) # Setup histories. V_last <- NULL loss.history <- matrix(NA, epochs + 1, attempts) error.history <- matrix(NA, epochs + 1, attempts) tau.history <- matrix(NA, epochs + 1, attempts) true.error.history <- matrix(NA, epochs + 1, attempts) dr <- cve(Y ~ X, k = k, method = method, momentum = momentum, epochs = epochs, attempts = attempts, logger = logger) # Plot history's matplot(loss.history, type = 'l', log = 'y', xlab = 'i (iteration)', main = paste('loss', name), ylab = expression(L(V[i]))) matplot(true.error.history, type = 'l', log = 'y', xlab = 'i (iteration)', main = paste('true error', name), ylab = expression(group('|', B*B^T - B[i]*B[i]^T, '|')[F] / sqrt(2*k))) matplot(error.history, type = 'l', log = 'y', xlab = 'i (iteration)', main = paste('error', name), ylab = expression(group('|', V[i-1]*V[i-1]^T - V[i]*V[i]^T, '|')[F])) matplot(tau.history, type = 'l', log = 'y', xlab = 'i (iteration)', main = paste('learning rate', name), ylab = expression(tau[i])) } cat("Created plot:", path, "\n")