monogenea

# Compute probabilities and predictions on test set
predictions <-  predict_classes(model, test_array)
probabilities <- predict_proba(model, test_array)

# Visual inspection of 32 cases
set.seed(100)
random <- sample(1:nrow(testData), 32)
preds <- predictions[random,]
probs <- as.vector(round(probabilities[random,], 2))

monogenea

# Fix structure for 2d CNN
train_array <- t(trainData$X)
dim(train_array) <- c(50, 50, nrow(trainData$X), 1)
# Reorder dimensions
train_array <- aperm(train_array, c(3,1,2,4))

test_array <- t(testData)
dim(test_array) <- c(50, 50, nrow(testData), 1)
# Reorder dimensions
test_array <- aperm(test_array, c(3,1,2,4))

monogenea

# Check processing on second cat
par(mar = rep(0, 4))
testCat <- t(matrix(as.numeric(trainData$X[2,]),
                    nrow = width, ncol = height, T))
image(t(apply(testCat, 2, rev)), col = gray.colors(12),
      axes = F)

# Save / load
save(trainData, testData, file = "catdogData.RData")
# load("catdogData.RData")

monogenea

# Takes approx. 15min
trainData <- extract_feature("train/", width, height)
# Takes slightly less
testData <- extract_feature("test1/", width, height, labelsExist = F)

monogenea

# Set image size
width <- 50
height <- 50

extract_feature <- function(dir_path, width, height, labelsExist = T) {
      img_size <- width * height
      
      ## List images in path
      images_names <- list.files(dir_path)
      

monogenea

##### Process image #####
library(keras)
library(EBImage)
library(stringr)
library(pbapply)

secondCat <- readImage("train/cat.1.jpg")
display(secondCat)

monogenea

# Validate on test set with ensemble
allPreds <- sapply(modelList, predict, newdata = testSet)
ensemblePred <- rowSums(allPreds) / length(modelList)

# Plot predicted vs. observed; create PNG
plot(ensemblePred, testSet$Y,
     xlim = c(0,100), ylim = c(0,100),
     xlab = "Predicted", ylab = "Observed",
     pch = 16, col = rgb(0, 0, 0, .25))
abline(a=0, b=1)

monogenea

bwplot(resamples(modelList),
       metric = "RMSE")

monogenea

# Train
doMC::registerDoMC(10)
knnMod <- train(myRec, data = trainSet,
                method = "knn",
                tuneGrid = data.frame(k = seq(5, 25, by = 4)),
                trControl = ctrl)

enetMod <- train(myRec, data = trainSet,
                 method = "glmnet",
                 tuneGrid = expand.grid(alpha = seq(0, 1, length.out = 5),

monogenea

# simple PCA, plot
pcaRec <- myRec %>% 
      step_pca(all_predictors())

myPCA <- prep(pcaRec, training = trainSet, retain = T) %>% 
      juice()
colGrad <- trainSet$Y/100 # add color

plot(myPCA$PC1, myPCA$PC2,
     col = rgb(1 - colGrad, 0, colGrad,.5),