从Seaborn Clustermap结果重新排序高级集群 [英] Reordering the high-level clusters from seaborn clustermap results

问题描述

是否可以通过脚本从下图中的a到b?我正在使用seaborn.clustermap()转到a(即，行的顺序得以保留.但是，列的顺序仅在第二高的级别上改变).

Is there a way to get from a to b in the following figure with scripting? I am using seaborn.clustermap() to get to a (i.e. the order of the rows are preserved. However, columns order change only at second highest level).

我想知道是否可以使用seaborn.clustermap()返回的seaborn.matrix.ClusterGrid，对其进行修改并绘制修改后的结果. b P.S..我要问的原因是订单具有含义(先是蓝色，然后是绿色，最后是红色).

I was wondering whether it is possible to use the seaborn.matrix.ClusterGrid that is returned by seaborn.clustermap(), modify it and plot the modified results. b P.S. The reason I am asking this is that the order has a meaning (first comes blue, next green, and finally red).

更新: 这是一个生成情况的小数据集:

Update: Here is a small data set to generate the situation:

df = pd.DataFrame([[1, 1.1, 0.9, 1.9, 2, 2.1, 2.8, 3, 3.1], 
                   [1.8, 2, 2.1, 0.7, 1, 1.1, 2.7, 3, 3.3]],
              columns = ['d1', 'd2', 'd3', 
                         'l3', 'l2', 'l1', 
                         'b1', 'b2', 'b3'],
              index = ['p1', 'p2'])

cg = sns.clustermap(df); ## returns a ClusterGrid

输出是这样的:

我们可以将以b开头的列视为早餐，将l开头的列用作午餐，将d开头的列视为晚餐.现在，顺序为breakfast -> dinner -> lunch.我想去breakfast -> lunch -> dinner.

We can think of columns starting with b as breakfast, l as lunch and d as dinner. Now, the order, is breakfast -> dinner -> lunch. I want to get to breakfast -> lunch -> dinner.

推荐答案

这就是我解决问题的方式.它可以工作，但可能不如人们所希望的那么优雅！

This is how I solved my problem. It works but it might not be as elegant as one would hope for!

import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
from scipy.cluster.hierarchy import linkage, dendrogram

# set the desired order of groups eg: breakfast, lunch, dinner
groups = ['b', 'l', 'd'] 

# reorder indexes/indices besed on the desired order
new_order = []
for group in groups:
    indexes = cg.data2d.columns.str.startswith(group)
    indexes_locs = np.where(indexes)[0].tolist()
    new_order += indexes_locs
    
## reorder df based on the new order
ordered_df = cg.data2d.iloc[:, new_order]

## Run clustermap on the reordered dataframe by disabling 
## the clustering for both rows and columns
ocg = sns.clustermap(ordered_df, 
                     row_cluster=False, 
                     col_cluster=False,
                    );

## draw dendrogram x-axis
axx = ocg.ax_col_dendrogram.axes
axx.clear()

with plt.rc_context({'lines.linewidth': 0.5}):
    
    link = cg.dendrogram_col.linkage ## extract the linkage information

    ## manualy inspect the linkage and determine the new desired order
    link[[4, 2]] = link[[2, 4]]  ## swaping the two groups of higher hierarchy
    
    ## draw the the dendrogram on the x-axis
    dendrogram(link, 
           color_threshold=0, 
           ax=axx,
           truncate_mode='lastp',
           orientation='top',
           link_color_func=lambda x: 'k'
          );

axx.set_yticklabels(['']*len(axx.get_yticklabels()))
axx.tick_params(color='w')
    
## draw dendrogram y-axis (no chage here)
axy = ocg.ax_row_dendrogram.axes
axy.clear()

with plt.rc_context({'lines.linewidth': 0.5}):
    
    ## draw the the dendrogram on the y-axis
    dendrogram(cg.dendrogram_row.linkage, 
           color_threshold=0, 
           ax=axy,
           truncate_mode='lastp',
           orientation='left',
           link_color_func=lambda x: 'k',
          );

axy.set_xticklabels(['']*len(axy.get_yticklabels()))
axy.tick_params(color='w')
# axy.invert_yaxis() # we might need to invert y-axis

输出看起来像这样:

The output looks like this:

这篇关于从Seaborn Clustermap结果重新排序高级集群的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！