Question

我想将虹膜数据集分为两个子集：一个用于执行PCA的子集（训练数据，包含物种“ versicolor”和“ setosa”），一个子集包含验证数据（包含物种“ virginica”） ”）。我想使用“ factoextra”软件包制作训练数据的PCA图，并向其中添加预测子集（验证数据）的点。

这是我到目前为止所拥有的：

library(dplyr)
library(tibble)
library(stringr)
library(factoextra)

scale_this = function(col){
  my_range = range(col)
  col = col - my_range[1]
  col = col / (my_range[2] - my_range[1])
  return(col)
}


my_data = iris

# Create annotation -----------------------------------------------------------
row_annotation = dplyr::select(my_data, Species)
row_annotation$Species = factor(row_annotation$Species)


# Create unique rownames + Scaling --------------------------------------------
my_data$unique_case_name = NA
for(i in 1:nrow(my_data)){
  my_data[i, "unique_case_name"] = paste0(i, "_", my_data[i, "Species"])
}

my_data = my_data %>% 
  dplyr::select(-Species) %>% 

  # scaling
  dplyr::mutate_at(vars(-unique_case_name), scale_this) %>%

  tibble::column_to_rownames(var = "unique_case_name") %>% 
  as.matrix()


# Separating the data ---------------------------------------------------------

# training: setosa, versicolor
training_indices = str_detect(string = rownames(my_data), pattern = "virginica", negate = T)
validation_indices = str_detect(string = rownames(my_data), pattern = "virginica", negate = F)


# Performing the PCA ----------------------------------------------------------
my_pca_object = prcomp(my_data[training_indices,],
                       scale. = F,
                       center = T,
                       retx = T)


# Predicting the validation data coordinates ----------------------------------
predicted_coordinates = stats::predict(my_pca_object, newdata = my_data[validation_indices,])


# Plot the PCA ----------------------------------------------------------------
# Plot the training data
my_pca_plot <- fviz_pca_ind(X = my_pca_object,
                            axes = c(1, 2),
                            geom.ind = c("point"),
                            addEllipses = T
) +
  labs(title = element_blank())

# Add validation data
my_pca_plot_2 <- fviz_add(my_pca_plot, predicted_coordinates,
                          color ="blue",
                          geom = c("point"),
                          addlabel = F,
                          shape = 8,
                          pointsize = 2)

my_pca_plot_2

我的问题： 1.我想在图中添加可变箭头：

我想按物种（万用色，setosa，弗吉尼亚州）为点着色
我希望在训练和验证子集的要点之间有视觉上的区别，例如不同的形状（但要保持物种的颜色）

我该如何解决这些问题？

如何用事实修改PCA图

0 个答案: