[ldd-ml] Address DMSP suggestions

[wkiri]

Describe the issue identified Address suggestions from DMSP review of this sub-model.

Describe the solution you'd like

• [x] Data_Set class: data_set_size could be misinterpreted as a byte size. Maybe data_set_count would be a better name.
• [ ] Data_Set class: Add an attribute called data_set_name or other attributes to document the source of the data_set. See comments for Training_Set, Validation_Set, Test_Set
• [x] Machine_Learning_Algorithm: algorthm_learning_style: For enumerated values, the meaning of labeled data is unclear. Explanation could be clearified in documentation.
• [x] Test_Performance: performance_score: It is not clear if the value has limits; is always positive; has a limit that means a perfect score. Can a min/max be added to definition and the description be revised to clarify?
• [x] Training_Set, Validation_Set, Test_Set: These (Data_Set class) classes do not include attributes to provide provence or processing details as noted in the class definitions. Add appropriate attributes to the Data_Set class.
• [x] Validation_Set, Test_Set: The difference between these two classes is not clear. Either update the class definitions and/or explain in the documentation.

Describe alternatives you've considered None

LDD Dictionary Version 1.0.1.0

PDS4 IM Version 1.18.0.0

Need-by Date When possible.

Additional context DMSP workshop follow-up.

[wkiri]

I've addressed all of the suggestions except the one to add a data_set_name attribute to the Data_Set class. Not all data sets have a designated name, so this might be hard to fill in. However, for provenance, it is worth thinking of a way to include a data_set_source (DOI?) which would give the option for a direct link to the original data set.

[jordanpadams]

@wkiri You could maybe leave this somewhat "open ended" and have a choice for how they want to reference the data set amongst a few options. Here are some that come to mind, but others may have some ideas:

• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#class_pds_internal_reference
• (DOI can be used here) https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#class_pds_external_reference
• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#class_pds_external_reference_extended

Not sure if these would be considered "source products" or not, but here are some others:

• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#class_pds_source_product_internal
• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#class_pds_source_product_external
• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#9.37%C2%A0%C2%A0class_pds_table_delimited_source_product_external
• https://pds.nasa.gov/datastandards/documents/im/v1/index_1I00.html#9.37%C2%A0%C2%A0class_pds_table_delimited_source_product_internal

[wkiri]

@jordanpadams Thanks! Yes, I think a pds.external_reference would be great. We use that for a DOI for the machine learning algorithm specification (when documented with a paper); see below. I think a DOI for the data set(s) likewise formatted would be a good addition.

https://github.com/pds-data-dictionaries/ldd-ml/blob/c13e6ca92c9bfc18b42778ad2f7bb06e9e17b67e/src/PDS4_ML_IngestLDD.xml#L388-L394