Missing Rows/Columns in Correlation Matrix with Duplicate Column Names After Name Shortening — No Warnings Raised

[guillaume-vignal]

Description:
I'm encountering a bug in Plotly where duplicate column names in a correlation matrix result in missing rows or columns when plotted using plotly.graph_objects.Heatmap. The key issue is that when there are columns with identical names, Plotly does not raise any warnings or errors, but silently drops some of the data, resulting in incomplete visualizations.

In my specific use case, I have a complex dataset with many columns. Due to the large number of features, some of the column names are quite long, which makes the axis labels in the heatmap difficult to read. To improve readability and create a cleaner plot, I shorten these column names programmatically. However, this shortening process can lead to multiple columns having identical labels (e.g., after truncating different names, they become the same).

The problem arises when these shortened column names are used in the correlation matrix:

• Plotly fails to handle the case of duplicate labels, causing entire rows or columns to disappear from the heatmap.
• More importantly, no warnings or errors are raised to inform me that the plot is incomplete. This lack of feedback makes it difficult to identify the root cause of the problem.

---

Steps to Reproduce:

1. Create a DataFrame with many columns, some of which have long names.
2. Shorten these column names for cleaner visualization.
3. Compute a correlation matrix using pandas.DataFrame.corr().
4. Plot the correlation matrix using plotly.graph_objects.Heatmap.

Example:

import pandas as pd
import numpy as np
import plotly.graph_objects as go

# Data with long column names (for demonstration)
data = {
    "Very_Long_Feature_Name_One": np.random.rand(100),
    "Another_Very_Long_Feature_Name_Two": np.random.rand(100),
    "Short_Feature_Three": np.random.rand(100),
    "Yet_Another_Very_Long_Feature_Name_Four": np.random.rand(100)
}

df = pd.DataFrame(data)

# Shorten the column names (as part of the process)
shortened_names = ['Feature_1', 'Feature_2', 'Feature_1', 'Feature_3']  # Simulating the shortening collision

# Replace columns with shortened names
df.columns = shortened_names

# Compute correlation matrix
corr_matrix = df.corr()

# Plot correlation matrix
fig = go.Figure(
    data=go.Heatmap(
        z=corr_matrix.values,
        x=corr_matrix.columns,
        y=corr_matrix.index,
        colorscale='Viridis'
    )
)

fig.update_layout(
    title="Correlation Matrix with Duplicate Shortened Names",
    xaxis_nticks=36
)

fig.show()

---

Expected Behavior:

1. Plotly should handle duplicate column names: Either by adding unique suffixes or indices to prevent label collisions or by providing an option to ensure that axis labels remain unique.
2. Warnings or Errors should be raised: If Plotly cannot handle duplicate labels, it should at least raise a warning or error to inform the user that duplicate labels exist and may cause issues in the plot. This would help users detect the problem early, especially in complex use cases.

---

Actual Behavior:

• Plotly drops some rows or columns in the correlation matrix when labels are identical.
• No warning or error is raised, leading to silent failures in the plot.
• The result is an incomplete correlation matrix, with missing rows or columns, and no indication that the plot is incorrect.

---

My Specific Use Case:

I am working with a large and complex dataset containing many columns, some with very long names. To create a visually appealing and readable heatmap, I shorten these column names. This is particularly important for clarity in reports or dashboards, where concise labels are preferable for aesthetics.

However, shortening the names leads to cases where different original feature names end up being shortened to the same label. For instance, "Very_Long_Feature_Name_One" and "Another_Very_Long_Feature_Name_Two" could both be shortened to "Feature_1". When this happens, Plotly seems unable to differentiate between the labels, causing rows and columns to go missing in the heatmap.

This issue becomes especially problematic because:

• I have no way of knowing that some rows or columns are missing unless I inspect the data closely.
• Plotly doesn't provide a warning or error when duplicate labels are present, leaving me unaware of the issue.

---

Environment:

• Plotly version: 5.24.1
• Python version: python 3.10
• OS: linux

---

Suggestions:

1. Handle Duplicate Labels Automatically:
   Plotly should have a built-in mechanism to automatically handle duplicate labels by adding a suffix (e.g., "_1", "_2") or provide options for handling label collisions.

2. Raise a Warning or Error:
   If Plotly detects that labels are not unique, it should raise a warning or error to inform the user, allowing them to address the issue before generating incomplete plots.

[turbotimon]

You may check if this is solved (at least for 2. Raise a Warning or Error) in version >=6 as something similar is done #3181