how to set tracking_metric_names dynamically

I have a function y=objective(x) which returns a dictionary of metrics y, one of which is the metric to optimize and the others are metadata. I understand I need to tell the experiment class to track those using tracking_metric_names. However, now every time I update objective(x) I need to update the variable tracking_metric_names. Is there a way to use the output of the first call to y=objective(x) to see the output keys and set the tracking_metric_names=[y.keys()] accordingly (after experiment was created)? I tried to read the source code but it seems not very accessible to do so, it’s not as simple as setting the class property.

Moreover, it’s not clear what metric_definitions param does, the documentation is very limited.

Thanks.

[danielcohenlive]

I’ve added the ability to do tracking metrics directly on the client. So now you can do

for i in range(25):
    parameters, trial_index = ax_client.get_next_trial()
    # Local evaluation here can be replaced with deployment to external system.
    result = evaluate(parameters)
    ax_client.complete_trial(trial_index=trial_index, raw_data=result)
    ax_client.add_tracking_metrics([k for k in result if k not in ax_client.metric_names])

I’m going to close this now, but feel free to ask further questions if you have them

[chanansh]

@danielcohenlive unfortunately it does not work:

AttributeError: ‘AxClient’ object has no attribute ‘metric_names’