Rename predicted_trajectory to reference_trajectory in ADK evaluation notebook

[prabha-git]

Description

Fixes a bug in gemini/evaluation/evaluating_adk_agent.ipynb where the evaluation dataset uses
the wrong column name, causing trajectory metrics to fail.

Fixes #2417

Problem

When running trajectory evaluation metrics (cell 43), the following error occurs:
Error: 400 List of found errors:
Field: trajectory_exact_match_input.instances[0].reference_trajectory
Message: Required field is not set.

Root Cause

Cell 28 creates the evaluation dataset with "predicted_trajectory" as the column name, but
Vertex AI Gen AI Evaluation service requires "reference_trajectory" for ground truth
comparisons.

Current code (cell 28):

eval_data = {
    "prompt": [...],
    "predicted_trajectory": [...]  # Wrong
}

Fixed code:
eval_data = {
    "prompt": [...],
    "reference_trajectory": [...]  # Correct
}

Solution

Renamed "predicted_trajectory" to "reference_trajectory" in cell 28's eval_data dictionary to
align with Vertex AI Gen AI Evaluation service requirements.

Testing

- Verified the notebook structure is preserved
-  Changed only the column name in the dataset definition (cell 28)
-  All trajectory metrics should now work correctly:
  - trajectory_exact_match
  - trajectory_in_order_match
  - trajectory_any_order_match
  - trajectory_precision
  - trajectory_recall

Impact

- Before: Users encounter 400 error when running trajectory evaluation
- After: Trajectory evaluation runs successfully without errors

Related Documentation

- https://cloud.google.com/vertex-ai/generative-ai/docs/models/evaluation-dataset

[gemini-code-assist]

Note

Gemini is unable to generate a summary for this pull request due to the file types involved not being currently supported.

[holtskinner]

@inardini Can you verify this?