Plot doesn't show on dashboard

Panel
1

I am running this code and all I get is white screen.

class refresh_dashboard(param.Parameterized):
    action = param.Action(lambda x: x.param.trigger('action'), label='Refresh')
       
    @param.depends('action')
    def get_plot(self):
        input_data = retention_table_plot()
        return plot_cohort_analysis(input_data)
    
cohort_analysis_dashboard = refresh_dashboard()

dashboard = pn.Column(
       pn.panel(cohort_analysis_dashboard.param, show_labels=True, show_name=False, margin=0),
       cohort_analysis_dashboard.get_plot, width=400
).servable()

dashboard
2

In which context do you run your code?

  • In a notebook you will have to add pn.extension() after having imported panel. This is necessary to prepare the notebook to display panel objects and have them being reactive.
  • As you script you’ll have to run it with panel serve yourscript.py --show
1 Like
3

Hi @cschweipert,

Do you get a button if you run this? It works for me, I did have to remove the code for the plot graph as it refers to stuff that isn’t present.

import panel as pn
import param
pn.extension()

class refresh_dashboard(param.Parameterized):
    action = param.Action(lambda x: x.param.trigger('action'), label='Refresh')
       
    @param.depends('action')
    def get_plot(self):
        return

cohort_analysis_dashboard = refresh_dashboard()

dashboard = pn.Column(
       pn.panel(cohort_analysis_dashboard.param, show_labels=True, show_name=False, margin=0))
dashboard

image

2 Likes
4

@maximlt @carl Thank you. I run it in a notebook. I added pn.extension(). Now the button shows but the graph doesn’t show in the browser. The graph shows in the notebook. It did show the button before sometimes even without pn.extension().

import os  
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import warnings
import seaborn as sns
import matplotlib.colors as mcolors
from operator import attrgetter
from sqlalchemy import create_engine

import operator as op
import param
import panel as pn
pn.extension()
5

Hi @cschweipert,

I think you would have to share your code here or at least a small a piece that can be run showing the problem for further help, looks like it might of been chopped

6

Hi @Carl
I rewrote it so it works with this sample dataset:
https://archive.ics.uci.edu/ml/machine-learning-databases/00502/online_retail_II.xlsx

def booking_data():
    df = pd.read_excel('Online Retail.xlsx',
                   dtype={'CustomerID': str,
                          'InvoiceID': str},
                   parse_dates=['InvoiceDate'], 
                   infer_datetime_format=True)
    return df

def retention_table_plot():
    """
    create retention matrix.
    """
    df = booking_data()

    df.dropna(subset=['CustomerID'], inplace=True)
    df['order_month'] = df['InvoiceDate'].dt.to_period('M') 
    df['cohort'] = df.groupby('CustomerID')['InvoiceDate'] \
                    .transform('min') \
                    .dt.to_period('M')
    df_cohort = df.groupby(['cohort', 'order_month']) \
                .agg(n_customers=('CustomerID', 'nunique')) \
                .reset_index(drop=False)
    df_cohort['period_number'] = (df_cohort.order_month - df_cohort.cohort).apply(attrgetter('n'))
    cohort_pivot = df_cohort.pivot_table(index = 'cohort',
                                        columns = 'period_number',
                                        values = 'n_customers')
    cohort_size = cohort_pivot.iloc[:,0]
    retention_matrix = cohort_pivot.divide(cohort_size, axis = 0)
    return [retention_matrix, cohort_size]

def plot_cohort_analysis(matrix):
    """
    plot retention table.
    """
    with sns.axes_style("white"):
        fig, ax = plt.subplots(1, 2, figsize=(40, 25), sharey=True, gridspec_kw={'width_ratios': [1, 11]})
    
        # retention matrix
        sns.heatmap(matrix[0], 
                    mask=matrix[0].isnull(), 
                    annot=True, 
                    fmt='.0%', 
                    cmap='YlGnBu', 
                    ax=ax[1], vmin=0, vmax=0.015)
        ax[1].set_title('Monthly Cohorts: User Retention', fontsize=20)
        ax[1].set(xlabel='# of periods',
                  ylabel='')

        # cohort size
        cohort_size_df = pd.DataFrame(matrix[1]).rename(columns={0: 'cohort_size'})
        white_cmap = mcolors.ListedColormap(['white'])
        sns.heatmap(cohort_size_df, 
                    annot=True, 
                    cbar=False, 
                    fmt='g', 
                    cmap=white_cmap, 
                    ax=ax[0])

        fig.tight_layout()

class refresh_dashboard(param.Parameterized):
    action = param.Action(lambda x: x.param.trigger('action'), label='Refresh')
       
    @param.depends('action')
    def get_plot(self):
        input_data = retention_table_plot()
        return plot_cohort_analysis(input_data)
    
cohort_analysis_dashboard = refresh_dashboard()

dashboard = pn.Column(
       pn.panel(cohort_analysis_dashboard.param, show_labels=True, show_name=False, margin=0),
       cohort_analysis_dashboard.get_plot, width=400
).servable()

dashboard
7

You need to return the figure in plot_cohort_analysis and instead of using plt.subplot you should use from matplotlib.figure import Figure

1 Like
8

Exactly, the docs explain why plt.subplot should not be used: Matplotlib — Panel 0.12.6 documentation

1 Like
9

That did not fix it. The graph still doesn’t show up in the browser.

10

What changes did you make? It’d be great if you could share your code.

11

This following code works for me. I think the main problem in your example is you are loading a lot of data which takes time and the plot cannot be shown before this is done. Therefore in my example I just removed that step and only import the already processed data in. This option is properly not available for you so you must find a way to speed up your data loading/processing, but this is definitely out of scope for this forum.

Another time you need help try to make a minimal, reproducible example (MRE). The excel file is 50 MB and took 5 min to download for me, the example could easily been done with fake data.

image
image1819Ă—1229 179 KB 

import matplotlib.colors as mcolors
import numpy as np
import pandas as pd
import seaborn as sns
from matplotlib.figure import Figure

import panel as pn
import param

pn.extension()
# run with panel serve $FILENAME.py

def retention_table_plot():
    df_cohort = pd.read_csv(
        "cohort.csv", index_col=0, parse_dates=["cohort", "order_month"]
    )

    cohort_pivot = df_cohort.pivot_table(
        index="cohort", columns="period_number", values="n_customers"
    )
    cohort_size = cohort_pivot.iloc[:, 0]
    retention_matrix = cohort_pivot.divide(cohort_size, axis=0)

    # Just to show the plot is updating
    retention_matrix.iloc[:, 1:] += np.random.rand() / 10

    return retention_matrix, cohort_size


def plot_cohort_analysis(matrix):
    """
    plot retention table.
    """
    with sns.axes_style("white"):
        fig = Figure(figsize=(13, 8))
        ax = fig.subplots(1, 2, sharey=True, gridspec_kw={"width_ratios": [1, 11]})

        # retention matrix
        sns.heatmap(
            matrix[0],
            mask=matrix[0].isnull(),
            annot=True,
            fmt=".0%",
            cmap="YlGnBu",
            ax=ax[1],
            vmin=0,
            vmax=1,
        )
        ax[1].set_title("Monthly Cohorts: User Retention", fontsize=20)
        ax[1].set(xlabel="# of periods", ylabel="")

        # cohort size
        cohort_size_df = pd.DataFrame(matrix[1]).rename(columns={0: "cohort_size"})
        white_cmap = mcolors.ListedColormap(["white"])
        sns.heatmap(
            cohort_size_df, annot=True, cbar=False, fmt="g", cmap=white_cmap, ax=ax[0]
        )

        ax[0].set_yticklabels(
            cohort_size_df.index.strftime("%Y-%m").to_list(), rotation=0
        )

        fig.tight_layout()

    return fig


class refresh_dashboard(param.Parameterized):
    action = param.Event(label="Refresh")

    @param.depends("action")
    def get_plot(self):
        input_data = retention_table_plot()
        return plot_cohort_analysis(input_data)


cohort_analysis_dashboard = refresh_dashboard()

dashboard = pn.Column(
    pn.panel(
        cohort_analysis_dashboard.param, show_labels=True, show_name=False, margin=0
    ),
    cohort_analysis_dashboard.get_plot,
    width=400,
).servable()

The cohort.csv

,cohort,order_month,n_customers,period_number
0,2009-12-01,2009-12-01,1045,0
1,2009-12-01,2010-01-01,392,1
2,2009-12-01,2010-02-01,358,2
3,2009-12-01,2010-03-01,447,3
4,2009-12-01,2010-04-01,410,4
5,2009-12-01,2010-05-01,408,5
6,2009-12-01,2010-06-01,408,6
7,2009-12-01,2010-07-01,374,7
8,2009-12-01,2010-08-01,355,8
9,2009-12-01,2010-09-01,392,9
10,2009-12-01,2010-10-01,452,10
11,2009-12-01,2010-11-01,518,11
12,2009-12-01,2010-12-01,260,12
13,2010-01-01,2010-01-01,394,0
14,2010-01-01,2010-02-01,86,1
15,2010-01-01,2010-03-01,119,2
16,2010-01-01,2010-04-01,120,3
17,2010-01-01,2010-05-01,110,4
18,2010-01-01,2010-06-01,115,5
19,2010-01-01,2010-07-01,105,6
20,2010-01-01,2010-08-01,91,7
21,2010-01-01,2010-09-01,114,8
22,2010-01-01,2010-10-01,134,9
23,2010-01-01,2010-11-01,122,10
24,2010-01-01,2010-12-01,37,11
25,2010-02-01,2010-02-01,363,0
26,2010-02-01,2010-03-01,109,1
27,2010-02-01,2010-04-01,82,2
28,2010-02-01,2010-05-01,110,3
29,2010-02-01,2010-06-01,93,4
30,2010-02-01,2010-07-01,76,5
31,2010-02-01,2010-08-01,79,6
32,2010-02-01,2010-09-01,103,7
33,2010-02-01,2010-10-01,100,8
34,2010-02-01,2010-11-01,106,9
35,2010-02-01,2010-12-01,32,10
36,2010-03-01,2010-03-01,436,0
37,2010-03-01,2010-04-01,95,1
38,2010-03-01,2010-05-01,113,2
39,2010-03-01,2010-06-01,103,3
40,2010-03-01,2010-07-01,100,4
41,2010-03-01,2010-08-01,87,5
42,2010-03-01,2010-09-01,105,6
43,2010-03-01,2010-10-01,130,7
44,2010-03-01,2010-11-01,126,8
45,2010-03-01,2010-12-01,36,9
46,2010-04-01,2010-04-01,291,0
47,2010-04-01,2010-05-01,67,1
48,2010-04-01,2010-06-01,58,2
49,2010-04-01,2010-07-01,47,3
50,2010-04-01,2010-08-01,54,4
51,2010-04-01,2010-09-01,67,5
52,2010-04-01,2010-10-01,79,6
53,2010-04-01,2010-11-01,76,7
54,2010-04-01,2010-12-01,22,8
55,2010-05-01,2010-05-01,254,0
56,2010-05-01,2010-06-01,49,1
57,2010-05-01,2010-07-01,45,2
58,2010-05-01,2010-08-01,49,3
59,2010-05-01,2010-09-01,48,4
60,2010-05-01,2010-10-01,66,5
61,2010-05-01,2010-11-01,56,6
62,2010-05-01,2010-12-01,22,7
63,2010-06-01,2010-06-01,269,0
64,2010-06-01,2010-07-01,58,1
65,2010-06-01,2010-08-01,53,2
66,2010-06-01,2010-09-01,55,3
67,2010-06-01,2010-10-01,62,4
68,2010-06-01,2010-11-01,76,5
69,2010-06-01,2010-12-01,20,6
70,2010-07-01,2010-07-01,183,0
71,2010-07-01,2010-08-01,38,1
72,2010-07-01,2010-09-01,37,2
73,2010-07-01,2010-10-01,52,3
74,2010-07-01,2010-11-01,55,4
75,2010-07-01,2010-12-01,21,5
76,2010-08-01,2010-08-01,158,0
77,2010-08-01,2010-09-01,39,1
78,2010-08-01,2010-10-01,50,2
79,2010-08-01,2010-11-01,51,3
80,2010-08-01,2010-12-01,20,4
81,2010-09-01,2010-09-01,242,0
82,2010-09-01,2010-10-01,73,1
83,2010-09-01,2010-11-01,63,2
84,2010-09-01,2010-12-01,28,3
85,2010-10-01,2010-10-01,379,0
86,2010-10-01,2010-11-01,112,1
87,2010-10-01,2010-12-01,39,2
88,2010-11-01,2010-11-01,322,0
89,2010-11-01,2010-12-01,38,1
90,2010-12-01,2010-12-01,47,0
12

@Hoxbro Thank you so much. The size of the dataset was the issue indeed. I just didn’t wait long enough. This is a fake sample dataset already, my real dataset is much bigger and gets fetched from a database. I didn’t know how to add a .csv file to this thread that’s why I used the sample dataset. Now, I know. Copy and paste. Should have been obvious. Thanks, again.

1 Like