[BUG] Confusion matrix visualizes target and predictions on wrong axes

[areshytko]

What steps did you take:

1. Run the following code:

from typing import NamedTuple
def conf_m() -> NamedTuple('conf_m_result', [('mlpipeline_ui_metadata', 'UI_metadata')]):
    import json
    from collections import namedtuple
    cm_data = 'False,False,22\nFalse,True,484\nTrue,False,12\nTrue,True,539\n'
    vocab = [False, True]
    metadata = {
        'outputs' : [{
            'type': 'confusion_matrix',
            'format': 'csv',
            'schema': [
                {'name': 'target', 'type': 'CATEGORY'},
                {'name': 'predicted', 'type': 'CATEGORY'},
                {'name': 'count', 'type': 'NUMBER'},
            ],
            'storage': 'inline',
            'source': cm_data,
            'labels': list(map(str, vocab)),
        }]
    }
    print(json.dumps(metadata))
    conf_m_result = namedtuple('conf_m_result', ['mlpipeline_ui_metadata'])
    return conf_m_result(json.dumps(metadata))

cm_op = kfp.components.func_to_container_op(conf_m)

@kfp.dsl.pipeline(name='test conf matrix')
def pipeline():
    op1 = cm_op()

kfp.Client().create_run_from_pipeline_func(pipeline, arguments={})

2. See result in KFP UI

What happened:

Axes were mixed up:

for the data:

What did you expect to happen:

Targets on the y-axis, predictions on the x-axis

Environment:

How did you deploy Kubeflow Pipelines (KFP)?

kubectl apply -k

KFP version: 0.5

KFP SDK version: 0.5.0

Anything else you would like to add:

[Miscellaneous information that will assist in solving the issue.]

/kind bug
/area frontend

[Bobgy]

/cc @numerology
I'm not familiar with the scope, is this the usual convention of showing a confusion matrix?
If yes, can you switch your schema order, to put predicted as the first item?

'schema': [
                {'name': 'target', 'type': 'CATEGORY'},
                {'name': 'predicted', 'type': 'CATEGORY'},
                {'name': 'count', 'type': 'NUMBER'},
            ],

[Bobgy]

@areshytko is above workaround good for you?

[areshytko]

Hey @Bobgy. Thanks for the reply.

The thing is not about the convention, the data on the plot is wrong: if you sum all targets on the plot you will get: 484+539 as True. And if you do it the original data in the table, you'll get: 539 + 12.

The workaround is confusing for the user because the order of columns in the data and in the provided metadata will differ.

[Bobgy]

@areshytko Thanks for the detailed explanation! I see, that's a bug indeed.