group_by with return_properties

[sebawita]

Describe your feature request

Hi there,

It would be great to run an aggregate query and return a selection of properties for matched objects within each group.

For example, if we have a collection of shopping items, like:

{
  "name": "red table",
  "price": 200,
  "type": "furniture",
  "description": "a beautiful red table with rounded corners like your iPhone"
}

It should be possible to search for "red furniture" and get back something like:

groups: [
  { 
     group: "red table", 
     objects: [ { name: ..., price: ...},  { name: ..., price: ...},  { name: ..., price: ...} ]
  },
  { 
     group: "ruby table", 
     objects: [ { name: ..., price: ...},  { name: ..., price: ...},  { name: ..., price: ...} ]
  },
]

Proposal

Can we add return_properties (or we could call it include_properties) to aggregate.near_x queries, like this:

response = artworks.aggregate.near_text(
    query="red furniture",
    object_limit=200,

    group_by=GroupByAggregate(
        prop="type", limit=3
    ),
    
    # proposal to add `return_properties`
    return_properties=[
        "name",
        "price",
        "type" # yes, it should be possible to return the group_by value as well
    ]
)

Then we should be able to access each group and their underlying objects (with the selected properties) like this:

for group in response.groups:
    for item in group.objects:
        print(item.properties) # access the requested properties
        print(item.id) # access the object id, it should be there by default

The current workaround

Currently, we can use:

return_metrics=Metrics("single_prop_name").text(),

or

return_metrics=[
    Metrics("name").text(),
    Metrics("description").text()
]

However, there is no way to get the required properties together - as in, we can't get the name+description together.

Also, the code access these properties is not the most elegant, and in this case we are not interested in how many times that value occurs:

for group in response.groups:
    for name in group.properties["name"].top_occurrences
        print(name.value)    

Code of Conduct

• I have read and agree to the Weaviate's Contributor Guide and Code of Conduct