ggplot2 热图 2 种不同的配色方案 - 混淆矩阵：不同配色方案中的匹配比错误分类

Question

我为来自 this answer.
的混淆矩阵改编了热图图但是我想扭曲它。在对角线上（从左上到右下）是匹配（正确的分类）。我的目标是在黄色调色板中绘制这条对角线。和红色调色板中的不匹配（所以除了对角线中的那些之外的所有瓷砖）。

在我的 plot.cm 函数中，我可以用

得到对角线

  cm_d$diag <- cm_d$Prediction == cm_d$Reference # Get the Diagonal
  cm_d$ndiag <- cm_d$Prediction != cm_d$Reference # Not the Diagonal

并且通过正确的 geom_tile 美学，我只能获得对角线（所需的黄色）配色方案

geom_tile( data = cm_d[!is.na(cm_d$diag), ],aes(color = Freq)) +
scale_fill_gradient(guide = FALSE,low=alpha("lightyellow",0.75), high="yellow",na.value = 'white')

但是我无法获得 cm_d$ndiag 元素的第二种配色方案我找到了包 ggnewscale that offers new_scale() as well as new_scale_fill().
I tired to implement it with the help of this blog。然而，对于热图的其余部分，结果只有深灰色填充的图块

# adapted from 
library(ggplot2)     # to plot
library(gridExtra)   # to put more
library(grid)        # plot together
library(likert)      # for reversing the factor order
library(ggnewscale)

plot.cm <- function(cm){
  # extract the confusion matrix values as data.frame
  cm_d <- as.data.frame(cm$table)
  cm_d$diag <- cm_d$Prediction == cm_d$Reference # Get the Diagonal
  cm_d$ndiag <- cm_d$Prediction != cm_d$Reference # Not the Diagonal     
  cm_d[cm_d == 0] <- NA # Replace 0 with NA for white tiles
  cm_d$Reference <-  reverse.levels(cm_d$Reference) # diagonal starts at top left

  # plotting the matrix
  cm_d_p <-  ggplot(data = cm_d, aes(x = Prediction , y =  Reference, fill = Freq))+
    scale_x_discrete(position = "top") +
    geom_tile( data = cm_d[!is.na(cm_d$diag), ],aes(color = Freq)) +
    scale_fill_gradient(guide = FALSE,low=alpha("lightyellow",0.75), high="yellow",na.value = 'white') +
    # THIS DOESNT WORK
    # new_scale("fill") +
    # geom_tile( data = cm_d[!is.na(cm_d$ndiag), ],aes(color = Freq)) +
    # scale_fill_gradient(guide = FALSE,low=alpha("red",0.75), high="darkred",na.value = 'white') +

    geom_text(aes(label = Freq), color = 'black', size = 6) +
    theme_light() +
    theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
      legend.position = "none",
      panel.border = element_blank(),
      plot.background = element_blank(),
      axis.line = element_blank())

  return(cm_d_p)
}

示例数据：
模拟插入符混淆矩阵

library(caret)
# simulated data
set.seed(23)
pred <- factor(sample(1:7,100,replace=T))
ref<- factor(sample(1:7,100,replace=T))
cm <- caret::confusionMatrix(pred,ref)
g <- plot.cm(cm)
g

Answer 1

我认为问题只是您指定的是 aes(color = Freq) 而不是 aes(fill = Freq。情节是你的目标吗？您还可以通过使用发散色标并创建一个新变量来简化所有这些，如果它不在对角线上，则将 Freq 标记为负数？请参阅下面的第二个示例

# adapted from 
library(ggplot2)     # to plot
library(gridExtra)   # to put more
library(grid)        # plot together
library(likert)      # for reversing the factor order
#> Loading required package: xtable
library(ggnewscale)

plot.cm <- function(cm){
  # extract the confusion matrix values as data.frame
  cm_d <- as.data.frame(cm$table)
  cm_d$diag <- cm_d$Prediction == cm_d$Reference # Get the Diagonal
  cm_d$ndiag <- cm_d$Prediction != cm_d$Reference # Not the Diagonal     
  cm_d[cm_d == 0] <- NA # Replace 0 with NA for white tiles
  cm_d$Reference <-  reverse.levels(cm_d$Reference) # diagonal starts at top left

  # plotting the matrix
  cm_d_p <-  ggplot(data = cm_d, aes(x = Prediction , y =  Reference, fill = Freq))+
    scale_x_discrete(position = "top") +
    geom_tile( data = cm_d[!is.na(cm_d$diag), ],aes(fill = Freq)) +
    scale_fill_gradient(guide = FALSE,low=alpha("lightyellow",0.75), high="yellow",na.value = 'white') +
    # THIS DOESNT WORK
    new_scale("fill") +
    geom_tile( data = cm_d[!is.na(cm_d$ndiag), ],aes(fill = Freq)) +
    scale_fill_gradient(guide = FALSE,low=alpha("red",0.75), high="red",na.value = 'white') +

    geom_text(aes(label = Freq), color = 'black', size = 6) +
    theme_light() +
    theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
          legend.position = "none",
          panel.border = element_blank(),
          plot.background = element_blank(),
          axis.line = element_blank())

  return(cm_d_p)
}

library(caret)
#> Loading required package: lattice
# simulated data
set.seed(23)
pred <- factor(sample(1:7,100,replace=T))
ref<- factor(sample(1:7,100,replace=T))
cm <- caret::confusionMatrix(pred,ref)
g <- plot.cm(cm)
g
#> Warning: Removed 8 rows containing missing values (geom_text).

^{由 reprex package (v0.3.0)}

于 2020-04-29 创建

# adapted from 
library(ggplot2)     # to plot
library(gridExtra)   # to put more
library(grid)        # plot together
library(likert)      # for reversing the factor order
#> Loading required package: xtable
library(ggnewscale)

plot.cm <- function(cm){
  # extract the confusion matrix values as data.frame
  cm_d <- as.data.frame(cm$table)
  cm_d$diag <- cm_d$Prediction == cm_d$Reference # Get the Diagonal
  cm_d$ndiag <- cm_d$Prediction != cm_d$Reference # Not the Diagonal     
  cm_d[cm_d == 0] <- NA # Replace 0 with NA for white tiles
  cm_d$Reference <-  reverse.levels(cm_d$Reference) # diagonal starts at top left

  cm_d$ref_freq <- cm_d$Freq * ifelse(is.na(cm_d$diag),-1,1)

  # plotting the matrix
  cm_d_p <-  ggplot(data = cm_d, aes(x = Prediction , y =  Reference, fill = Freq))+
    scale_x_discrete(position = "top") +
    geom_tile( data = cm_d,aes(fill = ref_freq)) +
    scale_fill_gradient2(guide = FALSE,low="red",high="yellow", midpoint = 0,na.value = 'white') +
    geom_text(aes(label = Freq), color = 'black', size = 6)+
     theme_light() +
    theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
          legend.position = "none",
          panel.border = element_blank(),
          plot.background = element_blank(),
          axis.line = element_blank())

  return(cm_d_p)
}

library(caret)
#> Loading required package: lattice
# simulated data
set.seed(23)
pred <- factor(sample(1:7,100,replace=T))
ref<- factor(sample(1:7,100,replace=T))
cm <- caret::confusionMatrix(pred,ref)
g <- plot.cm(cm)
g
#> Warning: Removed 8 rows containing missing values (geom_text).

^{由 reprex package (v0.3.0)}

于 2020-04-29 创建

ggplot2 热图 2 种不同的配色方案 - 混淆矩阵：不同配色方案中的匹配比错误分类

ggplot2 Heatmap 2 Different Color Schemes - Confusion Matrix: Matches in Different Color Scheme than Missclassifications

r

heatmap

ggplot2

confusion-matrix

r-caret