Enhancement: Expose the "truncate_mode" parameter of scipy.cluster.hierarchy.dendrogram() in seaborn.matrix.clustermap()

[YuanfengZhang]

1. The dendrogram is usually hard to read when the dataframe to draw the linkage is large, or it's too complex to show in the limited figure space sometime.
2. The workaround of preparing own linkage for the clustermap is kind of exhilarating and frustrating for users.

It would be very helpful for users drawing sns.clustermap with this parameter.
Thanks for your devotion and help.

Excepted:

sns.clustermap(*, truncate_mode: Optional[str] =None)

Todo:

1. add "self.truncate_mode" property at
   

   seaborn/seaborn/matrix.py

   Line 463 in 2c2c84e

   class _DendrogramPlotter:

2. add "self.truncate_mode" paramter at
   

   seaborn/seaborn/matrix.py

   Line 577 in 2c2c84e

   return hierarchy.dendrogram(self.linkage, no_plot=True,

3. add "self.truncate_mode" paramter at
   

   seaborn/seaborn/matrix.py

   Line 642 in 2c2c84e

   def dendrogram(

   
   

   seaborn/seaborn/matrix.py

   Line 687 in 2c2c84e

   plotter = _DendrogramPlotter(data, linkage=linkage, axis=axis,

   
   

   seaborn/seaborn/matrix.py

   Line 970 in 2c2c84e

   def plot_dendrograms(self, row_cluster, col_cluster, metric, method,

   
   

   seaborn/seaborn/matrix.py

   Line 974 in 2c2c84e

   self.dendrogram_row = dendrogram(

   
   

   seaborn/seaborn/matrix.py

   Line 984 in 2c2c84e

   self.dendrogram_col = dendrogram(

   
   

   seaborn/seaborn/matrix.py

   Line 1117 in 2c2c84e

   def plot(self, metric, method, colorbar_kws, row_cluster, col_cluster,

   
   

   seaborn/seaborn/matrix.py

   Line 1129 in 2c2c84e

   self.plot_dendrograms(row_cluster, col_cluster, metric, method,

   
   

   seaborn/seaborn/matrix.py

   Line 1146 in 2c2c84e

   def clustermap(

   
   

   seaborn/seaborn/matrix.py

   Line 1152 in 2c2c84e

   row_colors=None, col_colors=None, mask=None,

[mwaskom]

I am probably -1 on exposing individual parameters like this but don't immediately see why it wouldn't be possible to pass along a dictionary of such parameters. I feel like this has certainly been discussed before and I am actually a little surprised that this is not supported already which makes me think I'm missing something about why it would be complicated. Note that there would need to be two separate parameters for the row / col clustering to align with the rest of the interface.

Note that my real preference w/r/t clustermap is spinning it out of seaborn altogether as I don't think it fits well with the current direction of the library / API. But I've not found maintainers to adopt it.

[YuanfengZhang]

I am probably -1 on exposing individual parameters like this but don't immediately see why it wouldn't be possible to pass along a dictionary of such parameters. I feel like this has certainly been discussed before and I am actually a little surprised that this is not supported already which makes me think I'm missing something about why it would be complicated. Note that there would need to be two separate parameters for the row / col clustering to align with the rest of the interface.

Note that my real preference w/r/t clustermap is spinning it out of seaborn altogether as I don't think it fits well with the current direction of the library / API. But I've not found maintainers to adopt it.

Thanks for the clarification. I understand it and 'm closing this issue.