File:Uwot.png

Summary

# Prepare data
library(uwot)
library(ggplot2)
library(dplyr)

# training and test sets (as given)
iris_train <- iris[c(1:10, 51:60), ]
iris_test  <- iris[100:110, ]

# numeric feature matrices
x_train <- as.matrix(iris_train[, 1:4])
x_test  <- as.matrix(iris_test[, 1:4])

# Fit UMAP on training data (keep model)
set.seed(123)

umap_model <- umap(
  x_train,
  n_neighbors = 5,
  min_dist = 0.1,
  n_components = 2,
  ret_model = TRUE
)

# training embedding
train_embed <- as.data.frame(umap_model$embedding)
colnames(train_embed) <- c("UMAP1", "UMAP2")
train_embed$set <- "train"
train_embed$Species <- iris_train$Species

# Project test samples into training UMAP space
test_embed <- as.data.frame(umap_transform(x_test, umap_model))
colnames(test_embed) <- c("UMAP1", "UMAP2")
test_embed$set <- "test"
test_embed$Species <- iris_test$Species

# Combine and plot
plot_df <- bind_rows(train_embed, test_embed)

ggplot(plot_df, aes(x = UMAP1, y = UMAP2)) +
  geom_point(
    data = subset(plot_df, set == "train"),
    aes(color = Species),
    size = 3,
    alpha = 0.6
  ) +
  geom_point(
    data = subset(plot_df, set == "test"),
    shape = 21,
    fill = "red",
    color = "black",
    size = 4,
    stroke = 1
  ) +
  theme_minimal() +
  labs(
    title = "UMAP trained on iris_train with projected iris_test samples",
    subtitle = "Colored = training data (by Species), red circles = test data"
  )