Fix boxplot function KeyError

[aqlkzf]

Problem Description

The original boxplot function （experiments/RegInf/utils.py） was designed to create a box plot with marginal distributions. However, it contained a bug where the input DataFrame, data, was being overwritten after a groupby operation. This resulted in a loss of the original DataFrame structure, leading to a KeyError when attempting to group by the hue column later in the code. The function was trying to access a column that no longer existed in the modified DataFrame.

Proposed Changes

To resolve this issue, I have made the following changes to the function:

1. Separation of Data Manipulation and Plotting: I introduced a new variable, plot_data, as a copy of the input DataFrame. This ensures that the original data is not altered during the plotting process. All manipulations are performed on plot_data instead.

2. Categorical Conversion and Type Assertion: The function now checks and converts the x and hue columns to categorical data types if they are not already. This ensures that the grouping operations work as expected.

3. Stacked Bar Plot Calculation: The calculation for the fractions in the stacked bar plot has been corrected. Instead of modifying the original DataFrame, a new grouped_data variable is created. It stores the normalized value counts necessary for the stacked bar plot, preserving the original data.

4. Legend Handling: The function has been updated to include a legend only when there is more than one level in the hue column, enhancing the clarity of the plot when hue distinctions are present.

These changes correct the error and improve the function's robustness by preventing unintended side effects on the input data.

Additional Notes

The revised function has been thoroughly tested to ensure that it handles the input data correctly and that the resulting plots are generated as expected.

I believe these improvements will enhance the functionality and user experience for others utilizing this box plot function.