Add dashboard panels

How to add and configure monitoring panels.

By default, the monitoring dashboard for each new project is empty. You must define the dashboard composition in the code and select which metrics and test results to show. You can add multiple panels to the dashboard for each project.
Note: we plan to add the ability to add panels from the interface in the next releases.
Code example
Refer to the ML Monitoring QuickStart for a complete Python script that shows how to add several monitoring panels for a toy dataset.
Quickstart for ML Monitoring
How it works
Evidently snapshots contain multiple measurements. For example, when you log the DataDriftTable() Metric in a snapshot, it will contain the dataset drift summary, like this:
'number_of_columns': 15,
'number_of_drifted_columns': 5,
'share_of_drifted_columns': 0.3333333333333333,
'dataset_drift': False,
It will also contain data on individual column drift. Here is a partial example:
'column_name': 'age',
'column_type': 'num',
'stattest_name': 'Wasserstein distance (normed)',
'stattest_threshold': 0.1,
'drift_score': 0.18534692319042428,
'drift_detected': True,
You can visualize any individual measurements over time as long as you capture them inside snapshots. To do that, you must add panels to a monitoring dashboard inside a specific project. You must pass the chosen value (the MetricResult contained inside the logged snapshot) to specify what to plot.
For example, you might want to plot the share_of_drifted_columns, number_of_drifted_columns, or a drift_score for a specific column: all these measurements are available as a MetricResult inside the DataDriftTable() metric.
Add panel
To add a new panel to an existing project, use the add_panel() method. Here is an example of adding a new Counter type panel to show the share of drifting columns.
project.dashboard.add_panel(
    DashboardPanelCounter(
        title="Share of Drifted Features",
        filter=ReportFilter(metadata_values={}, tag_values=[]),
        value=PanelValue(
            metric_id="DatasetDriftMetric",
            field_path="share_of_drifted_columns",
            legend="share",
         ),
         text="share",
         agg=CounterAgg.LAST,
         size=1,
    )
)
Note: project.dashboard is an exemplar of the DashboardConfig class.
You can add multiple panels to the same project dashboard. They will appear in the order listed in the project.
Panel parameters
When you add a panel, you must specify its type and properties, such as width, title, etc. To define which measurements to visualize on a given panel, you will use the value parameter.
The sections below explains the available panel types and parameters.
Panel types
You can choose between the following panel types.
DashboardPanelCounter. It can include a single number with text or any custom text (this is a way to add a title to your dashboard).
DashboardPanelPlot. Displays any measurement as a line plot, bar plot, scatter plot or histogram. Here is the line plot example:
Class DashboardPanel
This is a base class for DashboardPanelPlot and DashboardPanelCounter. The parameters below apply to both types of panels.
Parameter
Description
id: uuid.UUID = uuid.uuid4()
Unique ID of the panel. Assigned automatically.
title: str

Usage:
title="My New Panel”
The name of the panel. It will be visible at the header of a panel on a dashboard.
filter: ReportFilter

metadata_values: Dict[str, str]
tag_values: List[str]

Usage:
filter=ReportFilter(metadata_values={}, tag_values=[])
Filters allow you to choose a subset of snapshots from which to display values on the panel.

To use filtering, you must provide metadata or tags when you log Reports or Test Suites.When you create a panel, you can reference these metadata_values or tag_values.
size: WidgetSize = WidgetSize.FULL

Available: 1, 2

Usage:
size=1
Sets the size of the panel:
1 for a half-width panel
2 for a full-sized panel (Default)
Class DashboardPanelCounter
DashboardPanelCounter allows you to add a Counter panel. You can also use this panel type to create text panels.
Example 1. To create a panel with the dashboard title only:
project.dashboard.add_panel(
    DashboardPanelCounter(
        filter=ReportFilter(metadata_values={}, tag_values=[]),
        agg=CounterAgg.NONE,
        title="Bike Rental Demand Forecast",
        )
    )
Example 2. To create a panel that sums up measurements (number of rows) over time.
project.dashboard.add_panel(
    DashboardPanelCounter(
        title="Model Calls",
        filter=ReportFilter(metadata_values={}, tag_values=[]),
        value=PanelValue(
            metric_id="DatasetSummaryMetric",
            field_path=DatasetSummaryMetric.fields.current.number_of_rows,
            legend="count",
        ),
        text="count",
        agg=CounterAgg.SUM,
        size=1,
    )
)
Parameter
Description
value: Optional[PanelValue] = None
The value (MetricResult) to show in the Counter.

You can create a simple text panel if you do not pass the Value.

See the section below on Panel Values for more examples.
text: Optional[str] = None
Supporting text to be shown on the Counter.
agg: CounterAgg

Available:
SUM, LAST, NONE
Data aggregation options:
SUM - sums the values from different snapshots over time
LAST - shows the last available value
NONE - to be used for text panels
Class DashboardPanelPlot
DashboardPanelPlot allows you to create scatter, bar, line, and histogram plots.
Example. To plot MAPE over time in a line plot.
project.dashboard.add_panel(
    DashboardPanelPlot(
        title="MAPE",
        filter=ReportFilter(metadata_values={}, tag_values=[]),
        values=[
        PanelValue(
            metric_id="RegressionQualityMetric",
            field_path=metrics.RegressionQualityMetric.fields.current.mean_abs_perc_error,
            legend="MAPE",
        ),
    ],
    plot_type=PlotType.LINE,
    size=1,
    )
)
Parameter
Description
values: List[PanelValue]
You must pass at least one value (MetricResult). You can also pass multiple values as a list. They will appear together: for example, as separate lines on a Line plot, bars on a Bar Chart, or points on a Scatter Plot. If you use a Histogram, the values will be aggregated.

See the section below on Panel Values for more examples.
plot_type: PlotType

Available: SCATTER, BAR, LINE, HISTOGRAM
Specifies the plot type.
Panel value
To add a numerical measurement to your panel, you must pass the PanelValue. For example, you can choose to display the number of drifting features, the share of empty columns, mean error, etc.
To define which exact value to show on a specific panel, you must specify the following:
A metric_id that corresponds to the Evidently Metric logged in a snapshot.
A field_path that corresponds to the specific MetricResult (numerical measurement) computed as part of this Metric. You can pass either a complete field path or a "field_name".
You can also pass the optional legend that will be visible on the plot.
Example 1. To include the share_of_drifted_columns MetricResult, available inside the DatasetDriftMetric():
value=PanelValue(
    metric_id="DatasetDriftMetric",
    field_path="share_of_drifted_columns",
    legend="share",
)
In this example, you pass the exact name of the field.
Example 2. To include the current.share_of_missing_values available inside the DatasetMissingValueMetric():
value=PanelValue(
    metric_id="DatasetMissingValuesMetric",
    field_path=DatasetMissingValuesMetric.fields.current.share_of_missing_values,
    legend="Missing Values Share",
)
In this example, you pass the complete field path inside the Metric
Note: you must always reference a metric_id, even if you used a Preset. For example, if you used a DataDriftPreset(), you can reference either of the Metrics it contains (DataDriftTable() or DatasetDriftMetric()). You can verify the Metrics included in each Preset here.
How to find the field path?
Option 1. Use autocomplete.
You can use autocomplete in interactive Python environments (like Jupyter notebook or Colab) to see available fields inside a specific Metric. They appear as you start typing the .fields. path for a specific Metric.
Note: some types of values (e.g. mean, sum, max, min) will not be visible using this method. This is because they match the names of the standard Python fields.
Option 2. Explore the contents of the snapshot, Metric or Test and find the relevant keys.
To look at all available measurements, you can also:
Open an existing snapshot file and explore its contents.
Generate a Report or a Test Suite, include the selected Metric or Test, and get the output as a Python dictionary. You can then explore the keys that contain the metric field names.
Once you identify the specific name of the field you would like to add to a panel, you can pass it as the field_path to the PanelValue parameter.
Note: in one of the next releases, we plan to add the ability to create panels from the visual interface and pre-built tabs with a fixed dashboard design. This will minimize the need to define the path to the metric values manually. If you need any help right now - ask in Discord!

Parameter	Description
`id: uuid.UUID = uuid.uuid4()`	Unique ID of the panel. Assigned automatically.
`title: str` Usage: `title="My New Panel”`	The name of the panel. It will be visible at the header of a panel on a dashboard.
`filter: ReportFilter` `metadata_values: Dict[str, str]` `tag_values: List[str]` Usage: `filter=ReportFilter(metadata_values={}, tag_values=[])`	Filters allow you to choose a subset of snapshots from which to display values on the panel. To use filtering, you must provide metadata or tags when you log Reports or Test Suites.When you create a panel, you can reference these `metadata_values` or `tag_values`.
`size: WidgetSize = WidgetSize.FULL` Available: `1`, `2` Usage: `size=1`	Sets the size of the panel: `1` for a half-width panel `2` for a full-sized panel (Default)

Parameter	Description
`value: Optional[PanelValue] = None`	The value (MetricResult) to show in the Counter. You can create a simple text panel if you do not pass the Value. See the section below on Panel Values for more examples.
`text: Optional[str] = None`	Supporting text to be shown on the Counter.
`agg: CounterAgg` Available: `SUM`, `LAST`, `NONE`	Data aggregation options: `SUM` - sums the values from different snapshots over time `LAST` - shows the last available value `NONE` - to be used for text panels

Parameter	Description
`values: List[PanelValue]`	You must pass at least one value (MetricResult). You can also pass multiple values as a list. They will appear together: for example, as separate lines on a Line plot, bars on a Bar Chart, or points on a Scatter Plot. If you use a Histogram, the values will be aggregated. See the section below on Panel Values for more examples.
`plot_type: PlotType` Available: `SCATTER`, `BAR`, `LINE`, `HISTOGRAM`	Specifies the plot type.

Log snapshots

Monitoring UI

Last modified 1mo ago