Python SDKSlackSign Up

Programmatically Create Dashboards

For most users, the Arize platform’s user interface (UI) offers the optimal experience for creating dashboards. The UI provides a live preview of your dashboard, making it easy to visualize changes as you add widgets. Moreover, with Arize’s customizable dashboard templates, you can quickly tailor dashboards to suit your specific requirements.

However, there are scenarios where programmatically creating dashboards might better serve your needs.If you aim to replicate a particular set of widgets across various dashboards, efficiently clone an entire dashboard, or integrate dashboard creation into your automated onboarding workflows, programmatic creation provides exceptional flexibility. This method allows you to handle custom and repeatable dashboard requirements effectively, providing a scalable solution for complex environments.

Creating a Dashboard Using a Template

Programmatically create a dashboard using one of our predefined templates, streamlining the setup process for common use cases

mutation create(
  $name:String!
  $modelId:ID!
  $template:DashboardTemplates!
  $postiveClass: String!
){
  createDashboardFromTemplate(input:
  {
    name:$name
    modelId:$modelId
    template: $template
    positiveClass: $postiveClass
  }
  ){
    dashboard{
      id
    }
  }
}

variables

{
  "name": "dashboard-name",
  "modelId": "model_id",
  "template": "templateName", 
  "postiveClass": "fraud"
}

Create an Empty Dashboard

When creating a new custom dashboard, you first need to create an empty state dashboard to add widgets to:

mutation createDashTest($name: String!, $spaceId: ID!) {
  createDashboard(input: { name: $name, spaceId: $spaceId }) {
    dashboard {
      id
    }
  }
}

This will return the dashboard id to use in mutations to add widgets.

Add a Distribution Widget

To visualize data variations and overall distribution patterns on your dashboard, use the createBarChartWidget mutation. 

mutation createBarChartWidget(
  $title:String!
  $dashboardId:ID!
  $plots:[BarChartPlotInputInput!]!
){
  createBarChartWidget(
    input:
  {
    title:$title
    dashboardId: $dashboardId 
    plots: $plots
  }
  ){
    barChartWidget{
      id
    }
  }
}

variable

{
  "title": "widget-title",
  "dashboardId": "dashboard_id",
  "plots": [
    {
      "title": "plot-title",
      "position": 1,
      "modelId": "model_id",
      "dimension": {
        "id": "dimension_id",
        "name": "dimension_name",
        "dataType": "STRING"
      },
      "dimensionCategory": "featureLabel",
      "filters": [],
      "modelVersionIds": []
    }
  ]
}

Add a Time Series Widget

Use the createLineChartWidget to add a time series widget to your dashboard, which is perfect for visualizing performance over time, data quality, and analyzing trends with options to add multiple lines or split data by specific criteria.

mutation createLineChartWidget(
  $title:String!
  $dashboardId:ID!
  $plots:[LineChartPlotInputInput!]!
  $timeSeriesMetricType: TimeSeriesMetricCategory!
){
  createLineChartWidget(
    input:
  {
    title:$title
    timeSeriesMetricType: $timeSeriesMetricType
    dashboardId: $dashboardId 
    plots: $plots
    widgetType: lineChartWidget
  }
  ){
    lineChartWidget{
      id
    }
  }
}

variables

{
  "title": "widget-name",
  "dashboardId": "model_id",
  "timeSeriesMetricType": "evaluationMetric",
  "plots": {
    "modelId": "model_id",
    "modelVersionIds": [],
    "dimensionCategory": "prediction",
    "metric": "auc",
    "positiveClass": "fraud",
    "title": "AUC Over Time",
    "position": 1,
    "modelEnvironmentName": "production",
    "filters": []
    }
}

Add a Drift Widget

To effectively monitor data drift, utilize the createLineChartWidget mutation with the widgetType input set to driftLineChartWidget. Below is an example that demonstrates using the PSI. You also have the flexibility to opt for other statistical methods such as Jensen-Shannon distance (JS) or Kullback-Leibler (KL) divergence, depending on your analysis needs.

mutation createDriftWidget(
$title:String!
  $dashboardId:ID!
  $plots:[LineChartPlotInputInput!]!
  $timeSeriesMetricType: TimeSeriesMetricCategory!
){
  createLineChartWidget(
    input:
  {
    title:$title
    timeSeriesMetricType: $timeSeriesMetricType
    dashboardId: $dashboardId 
    plots: $plots
    widgetType: driftLineChartWidget
  }
  ){
    lineChartWidget{
      id
    }
  }
}

variables

{
  "title": "drift-widget-title",
  "dashboardId": "dashbaord_id",
  "timeSeriesMetricType": "evaluationMetric",
  "plots": [{
    "modelId": "model_id",
    "modelVersionIds": [],
    "dimensionCategory": "prediction",
    "metric": "psi",
    "title": "drift-plot-title",
    "position": 1,
    "modelEnvironmentName": "production",
    "filters": [],
    "comparisonDatasetModelBaselineId": "baseline_id" 
  }
  ]
}

To get your baselineId you can query your model:

query getBaseline($modelID:ID!){
  node(id: $modelID) {
    ... on Model {
      modelPrimaryBaseline{
        id
      }
    }
  }
}

Add a Monitor Widget

Visualize your monitors using a monitor widget.

mutation createMonitorChartWidget(
  $title:String!
  $dashboardId:ID!
  $monitorId: ID!
){
  createLineChartWidget(
    input:
  {
    title:$title
    timeSeriesMetricType: evaluationMetric
    dashboardId: $dashboardId 
    monitorId: $monitorId
    widgetType: monitorLineChartWidget
  }
  ){
    lineChartWidget{
      id
    }
  }
}

variables

{
  "title": "widget-title",
  "dashboardId": "dashboard_id",
  "monitorId": "monitor_id"
}

Add a Statistic Widget

Add a statistics widget that displays key metrics, tailored to focus on either performance indicators or data quality metrics. This widget provides a quick overview and valuable insights at a glance.

mutation createStatWidget(
  $title:String!
  $dashboardId:ID!
  $timeSeriesMetricType: TimeSeriesMetricCategory!
  $modelId:ID!
  $performanceMetric: PerformanceMetric
  $positiveClass:String!
  $filters: [StatisticWidgetFilterItemInput!]!
){
  createStatisticWidget(
    input:
  {
    dashboardId: $dashboardId
    title:$title
    timeSeriesMetricType:$timeSeriesMetricType
    modelId:$modelId
    performanceMetric:$performanceMetric
    positiveClass: $positiveClass
    filters:$filters
    modelVersionEnvironmentMetadataIds:[]
    modelVersionIds:[]
  })
  {
    statisticWidget{
      id
    }
  }
}

variables

{
  "title": "widget-title",
  "dashboardId": "dashboard_id",
  "timeSeriesMetricType": "modelDataMetric",
  "modelId": "model_id",
  "aggregation":"DQ Metric",
  "dimension":{
    "id":"dimension_id"
    "name": "dimension_name",
    "dataType": "dimension_datatype"
  },
  "dimensionCategory": "prediction",
  "filters": []
}

Add a Text Widget

Add a text widget to your dashboard to offer explanatory text, headings, or contextual information, enriching the dashboard's narrative and usability.

mutation createTextWidget(
  $title:String!
  $dashboardId:ID!
  $content:String!

){
  createTextWidget(
    input:
  {
    dashboardId: $dashboardId
    title:$title
    content: $content
    creationStatus:published
    
  })
  {
    textWidget{
      id
    }
  }
}

variables

{
  "title": "text-widget-title",
  "dashboardId": "dashboard_id",
  "content": "content for the text widget"
}

Examples

Use case
Example Colab

Create a custom dashboard

Colab

Copy an existing dashboard

Colab

PreviousDriftNextPerformance Tracing

Last updated

Was this helpful?