[ENHANCEMENT][REFACTOR] SDK: allow to remove settings

[frascuchon]

Description

This PR allows users to change all the settings before creating the dataset, and metadata and vector, when the dataset is created.

I included the remove method instead of a new settings.add(override=True) since this change requires a lot of internal refactor with the current design.

An example of how to use could be to change inferred settings from hub before create the final dataset

settings = rg.Settings.from_hub("google/frames-benchmark")

settings.fields.remove("answer")

for field in settings.fields:
    if field.name.startswith("wiki"):
        settings.fields.remove(field)
        settings.metadata.add(rg.TermsMetadataProperty(field.name))

settings.questions.add(rg.TextQuestion(name="answer", title="Answer"))

dataset = rg.Dataset.from_hub("google/frames-benchmark", settings=settings)

Or adding new metadata or vector settings when the dataset is created:

dataset = client.dataset("my-dataset")

dataset.metadata.add([rg.TermsMetadata(name="split")])
dataset.update() # this line sends the change to the argilla server

Updated

You can remove a property by name:

settings.fields.remove("text")

or by property instance

for field in settings.fields:
   if field.name.startwith("wiki"):
      settings.fields.remove(field)

And you can override existing properties to change the property type, using the new settings.add method:

settings = Settings.from_hub(...)

# change some settings definitions before create the dataset
settings.add(rg.TermsMetadata("wikipedia_1")) # this will remove the existing wikipedia_1 property and will create a new terms metadata one

Type of change

• Refactor (change restructuring the codebase without changing functionality)
• Improvement (change adding some improvement to an existing functionality)

How Has This Been Tested

Checklist

• I added relevant documentation
• I followed the style guidelines of this project
• I did a self-review of my code
• I made corresponding changes to the documentation
• I confirm My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added relevant notes to the CHANGELOG.md file (See https://keepachangelog.com/)

[burtenshaw]

Just so I understand, there is still a singular way to remove settings? And this is just an example of iteration:

for field in settings.fields:
    if field.name.startswith("wiki"):
        settings.fields.remove(field)
        settings.metadata.add(rg.TermsMetadataProperty(field.name))

for example, fields can be deleted by their name:

settings.fields.remove("wiki_field")

[burtenshaw]

I included the remove method instead of a new settings.add(override=True) since this change requires a lot of internal refactor with the current design.

I get that this requires a refactor, but I think that with two steps, it makes the feature highly niche and expert.

Since we're working completely locally now, can we not add the settings.add(override=True) on top of this implementation, so that the add method round robins all attributes and tries to delete before adding the new attribute.

[frascuchon]

Just so I understand, there is still a singular way to remove settings? And this is just an example of iteration:

Yes, sorry. I didn't include it in the example. You can also remove the setting by name:

settings.fields.remove("text")

is a valid statement for this PR

[burtenshaw]

@frascuchon Should we put all the add and remove options in the description and the get feedback on the API?

[burtenshaw]

@frascuchon We haven't got feedback on this. I think we should move forward with the current implementation. Or ask again.

[frascuchon]

@frascuchon We haven't got feedback on this. I think we should move forward with the current implementation. Or ask again.

I will add some related tests and documents. I vote to merge it.

[frascuchon]

@frascuchon We haven't got feedback on this. I think we should move forward with the current implementation. Or ask again.

I will add some related tests and documents. I vote to merge it.

Done @burtenshaw. Here are the comments:

• tests (I changed the raised exception to SettingsError to align the rest of the code)
• docs

[frascuchon]

@burtenshaw take a look at the latest changes, please

[burtenshaw]

@frascuchon This looks good to me.

sorry for the delay.