Heatmap-Like Plot, But for Categorical Variables

heatmap-like plot, but for categorical variables

I decided it would be easist to approach this with ggplot2 (for me anyway):

#recreate a data set
dat <- data.frame(person=factor(paste0("id#", 1:50), 
    levels =rev(paste0("id#", 1:50))), matrix(sample(LETTERS[1:3], 150, T), ncol = 3))

library(ggplot2); library(reshape2)
dat3 <- melt(dat, id.var = 'person')
ggplot(dat3, aes(variable, person)) + geom_tile(aes(fill = value),
   colour = "white") + scale_fill_manual(values=c("red", "blue", "black"))

Write values in heatmap-like plot, but for categorical variables in seaborn

You can build the error texts and annotate manually:

c1, c2 = df_condition1.notna(), df_condition2.notna()
df_condition1,df_condition2 = df_condition1.fillna(''), df_condition2.fillna('')

errors = np.select((c1&c2, c1, c2), 
                   (df_condition1+'\n'+df_condition2, df_condition1, df_condition2),
                   '')

fig, ax = plt.subplots(figsize = (12, 10)) 
cmap = ['#b3e6b3','#66cc66','#2d862d','#ffc299','#ff944d','#ff6600','#ccddff','#99bbff','#4d88ff','#0044cc','#002b80']
ax = sns.heatmap(df, cmap=cmap, linewidths = 0.005, annot = False) 
    
for r in range(errors.shape[0]):
    for c in range(errors.shape[1]):
        ax.text(c+0.5,r+0.5, errors[r,c], 
                va='center',ha='center',
                fontweight='bold')

plt.show()

Output:

Heatmap-like plot for three categorical variables

How about this?

library(dplyr)
library(ggplot2)

df_max <- df %>%
  group_by(Color, Shape) %>%
  slice(which.max(Freq))

head(df_max)
# Source: local data frame [4 x 4]
# Groups: Color, Shape [4]
# 
#    Color    Shape   Size  Freq
#    (chr)    (chr)  (chr) (int)
# 1    Red   Square Medium     6
# 2    Red Triangle    Big    12
# 3 Yellow   Square    Big    10
# 4 Yellow Triangle  Small     8

ggplot(df_max, aes(x = Color, y = Shape, fill = Size)) +
  geom_tile()

Creating a heatmap based on extracted proportions out of a categorical data

After some data manipulation, you can use geom_tile:

library(tidyverse)

#Data wrangling
df1 <- 
  df1 %>% 
  group_by(AgeGroup, Sex) %>% 
  summarise(across(starts_with("marker"), 
                   ~ sum(.x == "yes") / n())) %>% 
  ungroup() %>% 
  mutate(gp = paste0(AgeGroup, "_", Sex), .keep = "unused") %>% 
  pivot_longer(-gp)

#   A tibble: 12 × 3
#   gp    name    value
#   <chr> <chr>   <dbl>
# 1 O_f   marker1 0.5  
# 2 O_f   marker2 1    
# 3 O_f   marker3 0    
# 4 O_m   marker1 0.5  
# 5 O_m   marker2 0.5  
# 6 O_m   marker3 0    
# ...

#Plot
df1 %>% 
  ggplot() +
  aes(x = name, y = gp, fill = value) +
  geom_tile() +
  theme_minimal()

plot a heatmap for binary categorical variables in R

You're on the right track with heatmap. Turn the "yes" / "no" columns of your df into a matrix of 0's and 1's and disable some of the defaults such as scaling and ordering.

mat1 <- 1*(df1[,-1]=="yes")

> mat1
     var1 var2 var3
[1,]    1    0    1
[2,]    1    1    0
[3,]    0    0    0
[4,]    1    1    1
[5,]    0    0    1

# You only need this step if you want the IDs to be shown beside the plot

rownames(mat1) <- rownames(df1)

> mat1
  var1 var2 var3
1    1    0    1
2    1    1    0
3    0    0    0
4    1    1    1
5    0    0    1

# reorder the matrix by rowSums before plotting

heatmap(mat1[order(rowSums(mat1)),], scale = "none", Rowv = NA, Colv = NA)

You can change the colour scheme by specifying the col parameter like

heatmap(mat1[order(rowSums(mat1)),], scale = "none", Rowv = NA, Colv = NA, col=c("lightgrey", "tomato"))

If you would prefer the plot to read left-to-right (one column per ID), just transpose the matrix

 heatmap(t(mat1[order(rowSums(mat1)),]), scale = "none", Rowv = NA, Colv = NA)

---

Related Topics