1 UMAP

Uniform Manifold Approximation and Projection (UMAP)

https://github.com/jlmelville/uwot

https://github.com/tkonopka/umap

2018年McInnes提出了算法，UMAP（Uniform Manifold Approximation and Projection for Dimension Reduction，一致的流形逼近和投影以进行降维）。 一致的流形近似和投影（UMAP）是一种降维技术，类似于t-SNE，可用于可视化，但也可用于一般的非线性降维。 该算法基于关于数据的三个假设：

  • 数据均匀分布在黎曼流形上（Riemannian manifold）；

  • 黎曼度量是局部恒定的（或可以这样近似）；

  • 流形是局部连接的。

https://jlmelville.github.io/uwot/index.html

Code
library(uwot)
head(iris)
#>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1          5.1         3.5          1.4         0.2  setosa
#> 2          4.9         3.0          1.4         0.2  setosa
#> 3          4.7         3.2          1.3         0.2  setosa
#> 4          4.6         3.1          1.5         0.2  setosa
#> 5          5.0         3.6          1.4         0.2  setosa
#> 6          5.4         3.9          1.7         0.4  setosa
colors = rainbow(length(unique(iris$Species)))
names(colors) = unique(iris$Species)
# umap2 is a version of the umap() function with better defaults
iris_umap2 <- umap2(iris[1:4]) |> as_tibble()
#> Warning: The `x` argument of `as_tibble.matrix()` must have unique column names if
#> `.name_repair` is omitted as of tibble 2.0.0.
#> ℹ Using compatibility `.name_repair`.

ggplot(iris_umap2,aes(V1,V2))+
    geom_text(aes(label=iris$Species),color=colors[iris$Species])

Code
# but you can still use the umap function (which most of the existing 
# documentation does)
iris_umap <- umap(iris[1:4]) |> as_tibble()

ggplot(iris_umap,aes(V1,V2))+
    geom_text(aes(label=iris$Species),color=colors[iris$Species])

Code
library(uwot)

set.seed(42) # 为了结果可重复
uwot_result <- umap(iris[1:4])

# 将结果转换为数据框
uwot_df <- as.data.frame(uwot_result)
colnames(uwot_df) <- c("UMAP1", "UMAP2")
uwot_df$Species <- iris$Species

# 可视化
ggplot(uwot_df, aes(x = UMAP1, y = UMAP2, color = Species)) +
  geom_point(size = 2) +
  labs(title = "UMAP of Iris Dataset (uwot)",
       x = "UMAP Dimension 1",
       y = "UMAP Dimension 2") +
  theme_minimal()

Back to top