Unhashable type: 'list' with MPL backend only

HoloViews
1

Hi.

I’m struggling to plot a graph using Matplotlib backend. It works fine with bokeh but I need to use matplotlib to save graphs as it is much faster.

I get a “TypeError: unhashable type: ‘list’” in /usr/local/lib/python3.6/dist-packages/holoviews/plotting/mpl/element.py

This is related to the Overlay object I am using. When I remove the labels it works.

Here is the stacktrace and plot function I use to plot the graph

combine_images
combine_images957×1348 297 KB

I use it with plot=False, and container.hv_backend as “matplotlib”
Are you able to tell me if I did something wrong or is this a bug ?

Thanks a lot !

2

We really need a reproducible example to help you. Can you paste the code you used?

3

Hi, thanks for your reply !
Since this code is part of a larger project, I cannot paste exactly the same code, but I have stripped down the code to create a snippet to reproduce error, you should be able to run it.
I tried to attach a csv file, but i’m not able to since I’m new, so here is a link to download it.
https://filebin.net/2jy5jyhpi35ut2hy/data.csv?t=jid1bnjr
And here is the code, you can test using matplotlib and bokeh by changing ‘back’, or test by saving or not using ‘plot’. I use save, but it does not seems to be the issue since the error arise both when plotting and saving

import pandas as pd
from typing import List, Union
import holoviews as hv
from holoviews.element.graphs import layout_nodes
import networkx as nx
from holoviews import opts, dim

def plot_graph(data: pd.DataFrame, features_columns: List[str], plot: bool=True, backend: str=None)-> Union[bool, hv.Overlay]:
    
    """
    plot a graph

    Args:
        data: (pd.DataFrame) dataframe
        cols_names: (List[str]) names of columns source, dest, value
        plot: (bool) wether to plot directly or return the plot object
        backend: (str) holoviews backend use to render
    Returns:
        if plot true:
            bool: True
        if plot false:
            hv.Overlay: the holoviews overlay object ready to plot
    """
    
    source, target, value = features_columns[0], features_columns[1], features_columns[2]
    
    df_links = data[[source, target, value]].rename(columns={source: "source", target: "target", value: "value"})
    df_nodes = pd.DataFrame({'name':pd.concat([df_links.source, df_links.target]).drop_duplicates()})
    
    nodes = hv.Dataset(df_nodes, 'name')
    
    as_graph = hv.Graph((df_links, nodes), ['source', 'target'], label="Airports flight connexions")
    
    as_graph = layout_nodes(as_graph, layout=nx.layout.fruchterman_reingold_layout)
        
    labels = hv.Labels(as_graph.nodes, ['x', 'y'], ['index'], label="airports")
    
    if backend == 'bokeh':
        (as_graph * labels).opts(
        opts.Graph(directed=True, node_size=10, bgcolor='white', xaxis=None, yaxis=None,
                   edge_line_color='black', edge_line_width=1, width=800, height=800, arrowhead_length=0,
                   node_fill_color='white', node_nonselection_fill_color='black'),
        opts.Labels(xoffset=-0.02, yoffset=0.02, text_font_size='8pt'))
    else:
        (as_graph * labels).opts(
        opts.Graph(directed=True, node_size=10, bgcolor='white', xaxis=None, yaxis=None,
                   edge_color='black', edge_linewidth=1, arrowhead_length=0),
        opts.Labels(xoffset=-0.02, yoffset=0.02, fontsize='8'))
    
        
    if plot:
        hv.output(as_graph*labels)
        return True
    return (as_graph*labels)

data = pd.read_csv("data.csv")

##variables##
plot = True
back = 'matplotlib'
#######

hv.extension(back)

graph = plot_graph(data, ["adep", "ades", "count"], plot=plot, backend=back)
if not plot:
    hv.save(graph, "graph.png", fmt='png')