11

I have 2 data tables with the dimensions 4x25. Each table is from a different point in time, but has exactly the same meta data, in essence the same column and row headers.

Given the large number of columns, I thought it best to represent this using a heatmap using the seaborn library for Python. However, I need to include both tables in the same plot. I am able to create a single heatmap representing a single data table as so.

df = pd.DataFrame(raw_data)
ax = sns.heatmap(df)
ax.set(yticklabels=labels)

However, I'm not sure how to combine two data tables into the same heatmap. The only way I can think of is to just create a new DataFrame of dimension 4x50 and then fit both tables into that one and plot that using the heatmap. But then, I need help with the following issues:

  1. I'm not sure how I'd draw a line down the middle of the heatmap to differentiate the data from the 2 tables.
  2. Another solution is to apply 2 different coloring schemes for the 2 sets of data within the same heatmap instead of simply just drawing a line down the middle.
Jonathan
  • 1,876
  • 2
  • 20
  • 56
  • 1
    I guess the reason you did not get an answer yet is also that you did not tag the question well. Seaborn is just a matplotlib extention, so any seaborn question should probably be tagged with matplotlib as well. The second reason is that you do not help people reproducing the issue. Hence if someone wanted to provide an answer they would need to replicate some example case themselves. Finally, there are a lot of possible ways to produce a plot as described. We do normally not want to write a compendium on each possible option. Instead ask a specific narrow question. – ImportanceOfBeingErnest Jan 17 '18 at 09:38
  • I see, thank you! I will definitely take that into account on my next question and update this one as well shortly. – Jonathan Jan 17 '18 at 20:28

3 Answers3

17

One possible way of showing two seaborn heatmaps side by side in a figure would be to plot them to individual subplots. One may set the space between the subplots to very small (wspace=0.01) and position the respective colorbars and ticklabels outside of that gap.

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns

df =  pd.DataFrame(np.random.rand(25,4), columns=list("ABCD"))
df2 = pd.DataFrame(np.random.rand(25,4), columns=list("WXYZ"))

fig, (ax,ax2) = plt.subplots(ncols=2)
fig.subplots_adjust(wspace=0.01)
sns.heatmap(df, cmap="rocket", ax=ax, cbar=False)
fig.colorbar(ax.collections[0], ax=ax,location="left", use_gridspec=False, pad=0.2)
sns.heatmap(df2, cmap="icefire", ax=ax2, cbar=False)
fig.colorbar(ax2.collections[0], ax=ax2,location="right", use_gridspec=False, pad=0.2)
ax2.yaxis.tick_right()
ax2.tick_params(rotation=0)
plt.show()

enter image description here

ImportanceOfBeingErnest
  • 321,279
  • 53
  • 665
  • 712
  • I've accepted this answer as it is a valid solution. But I imagined a different solution and I detailed that one in a question where it is easier for one to reproduce the data: https://stackoverflow.com/questions/48331097/nested-x-ticks-in-seaborn-heatmap – Jonathan Jan 18 '18 at 22:32
9

The best part about matplotlib/seaborn libraries is that everything is plotted in the same figure until you clear it. You can use the mask argument in sns.heatmap to get a diagonal heatmap plot. To get a "mixed" heatmap, such that you can have two different types of data plotted with different colormaps, you can do something like this:

from sklearn.datasets import load_iris
import seaborn as sns
import pandas as pd
import numpy as np

data = load_iris()
df= pd.DataFrame(data.data,columns = data.feature_names)
df['target'] = data.target

df_0 = df[df['target']==0]
df_1 = df[df['target']==1]

df_0.drop('target',axis=1,inplace=True)
df_1.drop('target',axis=1,inplace=True)

matrix_0 = np.triu(df_0.corr())
matrix_1 = np.tril(df_1.corr())

import seaborn as sns
from mpl_toolkits.axes_grid1.axes_divider import make_axes_locatable
from mpl_toolkits.axes_grid1.colorbar import colorbar
sns.heatmap(df_0.corr(),annot=True,mask=matrix_0,cmap="BuPu")
sns.heatmap(df_1.corr(),annot=True,mask=matrix_1,cmap="YlGnBu")

Hope this is what your second idea was. Note that this will only work when you have same column names.

Mixed Heatmap on Iris Dataset

Quark
  • 216
  • 2
  • 10
1

A slight twist of Quark's answer to avoid 0 values in the matrix that will cause those values showing the later cmap. We can compute boolean matrices to mask the upper/ lower triangle. More info here. Also added the limits of the cbars to fix the scale.

from sklearn.datasets import load_iris
import seaborn as sns
import pandas as pd
import numpy as np

data = load_iris()
df= pd.DataFrame(data.data,columns = data.feature_names)
df['target'] = data.target

df_0 = df[df['target']==0]
df_1 = df[df['target']==1]

df_0.drop('target',axis=1,inplace=True)
df_1.drop('target',axis=1,inplace=True)

mask_0 = np.zeros_like(df_0.corr(), dtype=np.bool_)
mask_0[np.tril_indices_from(mask_0)] = True
mask_1 = mask_0.T

import seaborn as sns
from mpl_toolkits.axes_grid1.axes_divider import make_axes_locatable
from mpl_toolkits.axes_grid1.colorbar import colorbar
sns.heatmap(df_0.corr(), annot=True, mask=mask_0, cmap="Blues", vmin=0, vmax=1)
sns.heatmap(df_1.corr(), annot=True, mask=mask_1, cmap="Greens", vmin=0, vmax=1)

2

Miroslav Afanasyev
  • 11
  • 1