Problem

When attempting to use Experiment.get_data_frame() with projects containing numerous runs, the API consistently returns 503 "Service Unavailable" errors. This appears to happen because the API attempts to fetch all experiment runs at once, which can overload the service when dealing with large datasets.

Current behavior

experiment = aiplatform.Experiment(experiment_name="specific-experiment-name")
df_experiment = experiment.get_data_frame(include_time_series=False)

This call fails with 503 errors when the experiment contains too many runs.

Proposed solution

Add date/time filtering parameters to the get_data_frame() method to limit the number of runs being processed:

df_run = run.get_data_frame(
    include_time_series=False,
    start_time="2023-01-01T00:00:00Z",
    end_time="2023-01-31T23:59:59Z"
)

This would allow users to retrieve manageable chunks of experiment data by time period, preventing service overload and improving reliability when working with large experiments.

Additional benefits

• Improved performance for large experiments
• Better user experience by avoiding timeout errors
• More control over data retrieval for analysis purposes
• Reduced load on backend services