visualize_ef.R

# #library(tidyverse)
# source("R/helperFunctions_ef.R")
# source("R/imageset_ef.R")
# source("R/image_ef.R")
# source("R/FeatureSpaceProjection.R")
#
# Import Olivetti

td <- load_imageset_ef("inst/extdata/olivetti_X.csv", c(64,64))
classes <- load_classes_ef("inst/extdata/olivetti_y.csv")

###############
# Plot example images


# Plot n examples of the Olivetti faces. 0 <= n <= 39.
plot_examples <- function(td, n, mode) {
  # n:    0 <= n <= 39.
  # mode: either ef or ex (for eigenfaces or examples)
  par(mfrow=c(round(sqrt(n)+0.49),round(sqrt(n)+0.49)),mar = c(2,2,2,2))

  if (deparse(substitute(mode)) == "ex") {
    for (i in 1:n) {
      idx <- i*10
      imgShow_ef(td[[idx]])
    }
    mtext("Example Images of the Olivetti Faces Dataset", side = 3, line = -1.5, outer = TRUE)
  }

  if (deparse(substitute(mode)) == "ef") {
    for (i in 1:n) {
      imgShow_ef(td[[i]])
    }
    mtext("Eigenfaces of the Olivetti Faces Dataset", side = 3, line = -1.5, outer = TRUE)
  }
  par(mfrow = c(1,1)) # Reset to original parameters
}

# How to use function
# This is the image used in README.md
#plot_examples(td, 30, mode=ef)


###############
# Plot eigenfaces

# Get eigenfaces and then reuse plot_examples function in mode=ef
ef <- get_eigenfaces(td, 9)
#ef
#plot_examples(ef,9, mode=ef)


################
# Reconstruction of faces

# TODO: turn this into nice looking format
# TODO: introduce argument. use n eigenfaces for reconstruction
#FSP(td[[5]], ef, avg_face(normalize(td)))
#td[[5]]

        ################
# Projection on two first eigenvectors
#library(ggplot2) #is in library(tidyverse)

ef1 <- as.vector(ef[[1]])
ef2 <- as.vector(ef[[2]])

# Initialization
data <- data.frame(matrix(ncol=3,nrow=5, dimnames=list(NULL, c("ef1_x", "ef2_y", "class"))))

# Fill with projections and corresponding classes
for (i in 1:length(td)) {
  data[i,] <- c(as.vector(td[[i]])%*%ef1,as.vector(td[[i]])%*%ef2,classes[[1]][i])
}

#data

# Scatterplot
#ggplot(data, aes(x=ef1_x, y=ef2_y, color=class)) + geom_point(shape=19, size=1.2) +
#geom_text(aes(label=class), size=2, hjust = -1, vjust = 0.3)
#  scale_color_gradientn(colours = rainbow(100))