Fits one multivariate Gaussian per class (quadratic discriminant analysis) to annotated pixels from one or several plates.
train_classifier(training, equal_priors = TRUE)Data frame from extract_training_pixels() (or a list
of them, which are row-bound).
Use equal class priors (recommended, because annotation effort rather than biology determines class frequencies).
An object of class mycohalo_classifier.
sim <- simulate_plate(width = 300, height = 400, seed = 5)
p <- read_plate(sim$image) |> detect_plate() |> model_background()
reg <- data.frame(class = c("fungus", "bacteria"),
x = c(sim$truth$colonies$x[1], p$dish$x),
y = c(sim$truth$colonies$y[1], p$dish$y), r = 8)
clf <- train_classifier(extract_training_pixels(p, reg))
clf
#> <mycohalo_classifier> Gaussian (QDA) pixel classifier
#> • fungus: n = 201; median |dE| = 33.4, colour direction (L,a,b) = (0.99, 0.05,
#> 0.089)
#> • bacteria: n = 208; median |dE| = 66.7, colour direction (L,a,b) = (0.98,
#> 0.015, 0.21)