Explainability
The Explainability tab lets you attach explainers to a trained model to understand how it makes decisions. DashAI supports two types of explainers: Global Explainers and Local Explainers.
Accessing Explainability
- Open a session from the left sidebar.
- Expand a model card that has a Finalizado status.
- Click the EXPLAINABILITY tab.
The tab is divided into two panels side by side:
| Panel | Description |
|---|---|
| Global Explainers | Analyze the model's behavior across the entire dataset — which features matter most overall. |
| Local Explainers | Analyze the model's behavior on individual predictions — why did the model produce a specific output for a specific input. |
Each panel shows a count badge of explainers already created, and a button to add a new one.
Global Explainers
Global explainers answer the question: "What does the model rely on in general?"
They produce results like feature importance rankings or SHAP summary plots that describe the model's overall decision-making patterns across the training or test data.
Adding a Global Explainer
Click + NEW GLOBAL EXPLAINER to open the explainer configuration flow.
The flow follows the same two-step pattern used throughout DashAI:
Step 1 — Configure Scope
Select which columns the explainer will analyze. The column selector table shows the index, name, value type, and data type of each column. Select the columns you want to include in the explanation.
Some explainers require a minimum number of columns — the counter at the top of the table shows how many are selected and the minimum required.
Step 2 — Configure Parameters
Each explainer has its own parameters. These vary by algorithm — common options include the number of samples to use, background dataset settings, and output format preferences. Each parameter has a ? help icon with a description.
Click CREATE EXPLAINER to run it and add the result to the tab.
Local Explainers
Local explainers answer the question: "Why did the model predict this specific output for this specific instance?"
They produce instance-level explanations — for example, showing which features pushed the prediction toward a particular class for a single row of data.
Adding a Local Explainer
Click + NEW LOCAL EXPLAINER to open the explainer configuration flow.
The configuration follows the same two-step pattern as global explainers, with scope selection (columns) and parameter configuration.
Some local explainers additionally require you to select a specific prediction or row to explain — this varies by the explainer type.
Explainer Results
Once created, each explainer appears as a result card in its panel with:
- The explainer name and type.
- A Finalizado badge when complete.
- The explanation output rendered inline (chart, table, or visualization depending on the explainer type).
- An Edit button to modify scope or parameters and re-run.
- A Delete button to remove the explainer.
Tips
- Run a Global Explainer first to get an overview of feature importance — this helps identify which input columns matter most and which may be redundant.
- Use Local Explainers when a specific prediction seems unexpected — they help trace which features drove an unusual result.
- Explainability tools are available per model — you can run the same explainer on multiple models in the same session to compare how different algorithms use the same features.
- The EXPLAINABILITY tab badge shows a count of created explainers, making it easy to track which models have been analyzed.
Troubleshooting
| Symptom | Likely cause | Solution |
|---|---|---|
| EXPLAINABILITY tab is inactive | Model has not been trained | Train the model first before adding explainers |
| Explainer creation fails | Incompatible column types for the selected explainer | Review the explainer's column type requirements and adjust the scope |
| No results rendered after creation | Processing error | Check the Job Queue for error details and retry |