Usage

See ./notebooks/examples.ipynb for example usage.

Functions:

create_error_profile(df,
                    id,
                    columns,
                    set_annotations,
                    incorrect_value,
                    figsize)

create_oddratio_profile(df,
                    subgroup,
                    id,
                    columns,
                    set_annotations,
                    incorrect_value,
                    figsize)

create_stratified_error_profile(df,
                    subgroup,
                    id,
                    columns,
                    set_annotations,
                    incorrect_value,
                    figsize)

• df: Dataframe where columns represent correct/incorrect predictions from different models
• id: column representing identifiers for each row/sample
• columns: which columns represent models
• set_annotations: names of each model
• incorrect_value: value in column representative of an error
• subgroup: additional categorical column stratifying samples into subpopulations

Example

Below is an example dataframe.

display(df.head())

	Model 1	Model 2	Model 3	Model 4	Model 5	Patient ID	Subgroup
0	1	0	1	1	1	1	A
1	0	1	0	1	0	2	B
2	0	0	0	0	0	3	B
3	0	0	1	0	0	4	A
4	0	0	0	0	0	5	A
...

create_error_profile(
    df,
    id="Patient ID",
    columns=[f"Model {i + 1}" for i in range(models)],
    incorrect_value=1,
)

Credit

Source repository: https://github.com/gecko984/supervenn. Most of the algorithms are adapted from the original repo. The edits are to extend the visualizations for error analysis.