feat: include ValidMeasurement metadata in output of ado show entities

[VassilisVassiliadis]

Is your feature request related to a problem? Please describe.

There are cases in which an actuator sets the value of an observed property and then provides additional information in the metadata section of the associated ValidMeasurement structure. However, this information is not exported to end-users when working with ado in a way that the following information is present in a tabular form:

1. the values of the entityspace for the measurement
2. the experiment identifier
3. the observed properties of the measurement
4. the additional information from the ValidMeasurement metadata

Currently, storing the above information in a single CSV file requires multiple calls to the ado api and potentially writing custom python code that combines their output into a single CSV file.

For example, the experiments of the SFTTrainer actuator mark invalid configurations by setting the value of the observed property is_valid to 0. There are multiple reasons for a configuration to be invalid and it is complicated to discern what caused the configuration to be tagged as invalid.

Describe the solution you'd like

I would like a way to include select fields of the metadata section within a ValidMeasurement to the output of ado show entities.

Describe alternatives you've considered

An actuator could record the additional information in new observed properties instead of using the ValidMeasurement metadata section.

For example, the SFTTrainer could produce an additional observed property called invalid_reason for measurements which have is_valid=0. The value of invalid_reason is one of the following:

• GPU_OOM
• BATCHSIZE_NOT_DIVISIBLE_BY_GPUS
• GPUS_NOT_DIVISIBLE_BY_FASTMOE
• EXPERTS_NOT_DIVISIBLE_BY_FASTMOE

This is misusing what an observed property should record and is thus against best practices.

Additional context
N/A

[michael-johnston]

Can you give a ado show entities command line to illustrate how you would use it?