Skip to Content
BayesiaLabBayesiaLab User GuideInferenceAutomatic Evidence-Setting

Automatic Evidence-Setting

Context

  • Automatic Evidence-Setting is a function that takes single observations from a dataset or sets of evidence from an Evidence Scenario File and performs inference with the current Bayesian network.
  • So, with Automatic Evidence-Setting, you can inspect — record-by-record — how each individual set of observations/evidence relates to the values inferred through the Bayesian network.
  • If the Bayesian network features a Target Node, Automatic Evidence-Setting compares the inferred values to the observed states.
  • In earlier versions of BayesiaLab, Automatic Evidence-Setting was known as Interactive Inference.

Usage

  • To start Automatic Evidence-Setting, select Main Menu > Inference > Automatic Evidence-Setting.
  • If both an Evidence Scenario File and an associated dataset are available for inference, BayesiaLab prompts you to choose your source:
AutomaticEvidenceSettingEvidenceSource
  • Once Automatic Evidence-Setting is activated, an additional control panel appears in the Toolbar.
ToolbarAutomaticEvidenceSetting

  • This control panel allows you to scroll through the associated dataset or Evidence Scenario File, depending on what you selected.

    • Clicking the button takes you back to the first record, i.e., record 0 in the dataset or scenario 0.

    • The button selects the previous record/scenario.

    • The counter box shows the record/scenario number of the currently displayed set of evidence.

    • The button selects the next record in the dataset/scenario.

    • Clicking takes you to the last record in the dataset/scenario.

    • Click the button to include Not-Observable Nodes for inference, even though they are normally excluded.

    • If you are using an Evidence Scenario File for inference, you can select a specific scenario by right-clicking on the counter box in the control panel. A dropdown list shows all available scenarios in the Evidence Scenario File.

    ControlPanelCounterEvidenceScenario
    • If you are using a dataset that contains a Row Identifier, you can search for a specific record by right-clicking on the counter box in the control panel and entering a search string. You can also use wildcards, i.e., ? to represent a single character and * for any number of characters. Upon pressing Enter, you obtain a list of results, from which you can select a specific record for inference.
    ControlPanelCounterRowIdentified

    • If you scroll through the records associated with a predictive Bayesian network model featuring a Target Node, you can compare the observed states of the Target Node with the states inferred or predicted by the model.

    • For instance, the following Monitors show the observed states for all predictor nodes (highlighted in green) plus the corresponding inferred distribution of the Target Node diagnosis.

    • Here, the inferred probabilities are approximately 75% for Malignant and 25% for Benign. The bar highlighted in silver indicates the actual observation in this particular case, i.e., the ground truth, which is that the sample was indeed Malignant.

      Loading SVG...
      Click to Zoom
      Loading SVG...

    • Going through records this way, comparing prediction vs. actual states one by one, can help you develop intuition for the sometimes abstract performance measures generated through Network Performance Analysis.

    • Clicking exits the Automatic Evidence-Setting mode.

Workflow Illustration

Approximate InferenceBatch Inference