[Feature Request]: add batch-fetch of entitites to ServingClient

[Anntey]

Description

• Recommender systems utilise feature stores typically
• Ranking models are a subset of recommender models commonly used in production
• Ranking models create an N-to-N mapping (where as retrieval models create an 1-to-N mapping). This means that you need to fetch n_features for N entities per recommendation request.
• Ie. first you fetch features feat_1, feat_2, feat_3 for 1 customer: result is 1 x n_cust_features. Second, you predict with model for each content entity for that single customer. This means that you fetch a set of features for all those entitites: result is n_content x n_cont_features. NOTE: this is not the same as "fetch features for all entitites" as the list of entity ids varies.

Current Behavior

result = []

for id in entity_ids:
    fpf = client.features(
        features = [
            ('content_feat_1', 'default'),
            ('content_feat_2', 'default')
        ],
        entities = {'content': id}
    )

    result.append(fpf)

Desired Behavior

fpf = client.features(
    features = [
        ('content_feat_1', 'default'),
        ('content_feat_2', 'default')
    ],
    entities = {'content': ['uid_1', 'uid_2', ..., 'uid_75']}
)

Returns

{'uid_1': [feat_1_value, feat_2_value], 'uid_2': [feat_1_value, feat_2_value], ..., 'uid_75': [feat_1_value, feat_2_value]}
or
([feat_1_value_1, feat_1_value_2, ..., feat_1_value_75], [feat_2_value_1, feat_2_value_2, ..., feat_2_value_75])

I suppose the underlying command for eg redis would look like

redis> HMGET content_feat_1 uid_1 uid_2 ... uid_75
redis> HMGET content_feat_2 uid_1 uid_2 ... uid_75

Benefits

This is MUST-HAVE feature for any organisation that uses Featureform for recommender systems.

Possible Implementation

No response

Additional Context

No response

[Anntey]

Isnt this batch_features() a "get all from index" basically https://github.com/featureform/featureform/blob/main/client/src/featureform/serving.py#L160. Which is different than fetching a subset.