Klarity

Appearance

Confusion_matrix ​

Used to display the performances of a classification model when applied on dataset

Instances ​

Only one instance of the confusion matrix is available right now, it uses two different graphical component depending on the abstraction level.

Configuration ​

Type configuration with specific parameter that need to be describe

JSON
   ./config.json
   
   "metrics": [
    {
      "name": "CONFUSION_MATRIX",
            "scope": "datasetModel",
            "type": "confusion_matrix",
            "title": "Confusion matrix",
            "description": "Confusion matrix description",
            "tooltip": "Tooltip operation roi",
            "config": {
                "horizontal_list": [
                    "Predicted",
                    [
                        "Positive",
                        "Negative",
                    ]
                ],
                "vertical_list": [
                    "Ground Truth",
                    [
                        "Positive",
                        "Negative",
                    ]
                ]
            }
        },

Payload ​

JSON

    ./payload.json

    
        {
            "metricName": "CONFUSION_MATRIX",
            "values": [
                [
                    8345,
                    35,
                    10
                ],
                [
                    27,
                    235,
                    5
                ],
                [
                    2,
                    2,
                    2
                ]
            ]
        }

Representations ​

Company Level representation ​

Two different instances :

!Alt text(images/image-3.png)

Color scheme for the different gradients :

!Alt text(images/image-3.png)

Cells depicted in white will follow the top gradient, cells depicted in grey will follow the bottom gradient.

!Alt text(images/image-3.png)

Project Level representation ​

!Alt text(images/image-3.png)

Color Ratio Diagonal = Green Others = Reddish Total per line = Blue

MCC Ratio = red to green. MCC Ratio formula Alt text

InDepth Level representation ​

alt text

  • clicking one of the cell (see the mouse cursor) will display content in a sample viewer at the bottom of the matrix. This sample is of course made of picture/elements from the selected cell.
Comparing with multiple models ​

Clicking the diff/compare option in the toolbar will open a list with checkboxes containing the different version that can be compared with the current one.

alt text

When comparison is active, each cell will display :

  • in top left corner the difference with the smallest value found for this cell among all the compared versions.

  • in top right coner the difference with the highest value found for this cell among al the compared versions.

  • it is displayed in red if the difference indicates the result is worst than the actual version.

  • it is displayed in green if the difference reveals a better result than the actual version.

    • (E.G. here, on the cell 8345, among all the compared models, the model with the lowest value for this cell has a value of 8343 which is the (-2) indicated in top left corner. The model with the highest value has 8389 which is the (+44) indicated in top right corner.
    • as the ground truth value here is 8390, it means the 8389 is closer to the ground truth value, though the differnce appears in green.

Navigating over a Diff will display an infobox with the associated version, the cell value and the diff with the actual version.