SNOW-2111386: Feature Store - Duplicate Feature Names in Feature Views Cause Issues with Generated Datasets

[ArtemioPadilla]

Hi,

I am experiencing an issue when using multiple feature views with features that share the same name, such as avg_balance, which leads to confusion in the final dataset. Here’s the scenario:

I have a spine dataframe with the following structure:

| user_id | state_id | postal_code | timestamp |

Additionally, I have three feature views corresponding to three entities: user, state, and postal_code. Each feature view contains a feature named avg_balance. When I create a dataset from these feature views, the resulting dataframe includes multiple columns with the same name (avg_balance), which makes it difficult to work with the data.

Current behavior:
The generated dataset has duplicated feature names (e.g., avg_balance), causing issues when trying to manipulate or analyze the data.

Expected behavior:
It would be helpful to add an option to automatically append a prefix or suffix to the feature names in the generated dataset to avoid these naming conflicts.

For example:

user_avg_balance

state_avg_balance

postal_code_avg_balance

This would help clarify the origin of each feature and make the dataframe easier to work with.

Suggestion:
Could an option be implemented to automatically add a prefix or suffix to features with the same name when merging datasets from multiple feature views?

Thank you for your support!

[sfc-gh-okharatsidi]

Hi there! It is indeed a limitation at the moment. It will be addressed at some point, but for now, the only workaround I can offer is making sure the feature names are unique across the feature views.

[sfc-gh-sfield]

Another option, rather than directly adjust the feature-names in the FeatureViews, is to create a downstream FeatureView for consumption, that is implemented as a View (refresh_freq = None), renaming any columns as needed. You can then use these downstream FeatureView's for the retrieval and avoid the column-name clashes. By sourcing the dataframe from the fully-qualified name of the original FeatureView you can create a view that references the underlying featureview database object

e.g.

session.sql(f'''
create or replace table {sess_db}.{fs_name}.user_raw 
(
user_id integer,
avg_balance decimal(12,2),
valid_from_ts timestamp
)
as
select
user_id,
avg_balance,
valid_from_ts
from (values
(1, 10.0, '2025-05-28 00:00:00'),
(1, 11.0, '2025-05-28 00:00:00'),
(1, 12.0, '2025-05-28 00:00:00'),
(2, 20.0, '2025-05-28 00:00:00'),
(2, 21.0, '2025-05-28 00:00:00'),
(2, 22.0, '2025-05-28 00:00:00')
) as u (user_id, avg_balance, valid_from_ts)
''')

# Create User Entity
ENT_USER = Entity(
    name="USER",
    join_keys=["USER_ID"],
    desc="User entity"
    )
fs.register_entity(ENT_USER)

# Create Raw User FeatureView in Feature Store
FV_USER_RAW = FeatureView(
    name = f"FV_USER_RAW",
    entities = [ENT_USER],
    timestamp_col = 'VALID_FROM_TS',
    feature_df = session.table(f'{sess_db}.{fs_name}.USER_RAW'),
    refresh_freq = '1 days',
    desc = f"User raw features from source"
)
# Register the User FeatureView in the schema, and add the python featureview to our list
FV_USER_RAW_V1 = fs.register_feature_view(
    feature_view = FV_USER_RAW, 
    version = "V1"    
)

fs.suspend_feature_view(FV_USER_RAW_V1)

# Helper function to rename all features in a FeatureView with prefix/suffix and return dataframe
def rename_features (fv, prefix = '', suffix = ''):
    fv_df = session.table(fv.fully_qualified_name())
    fcols = list(map(str, fv.feature_names))
    return (fv_df.with_columns([ prefix + item + suffix  for item in fcols], 
                         [ F.col(item)  for item in fcols] )
                .drop(fcols)
               )

# Create User Consume FeatureView in Feature Store
FV_USER_CONSUME = FeatureView(
    name = f"FV_USER_CONSUME",
    entities = [ENT_USER],
    timestamp_col = 'VALID_FROM_TS',
    feature_df = rename_features(FV_USER_RAW_V1, 'UR_'),
    refresh_freq = None,
    desc = f"User features for consumption from source"
)
# Register the User Consume FeatureView in the schema, and add the python featureview to our list
FV_USER_CONSUME_V1 = fs.register_feature_view(
    feature_view = FV_USER_CONSUME, 
    version = "V1"    
)

FV_USER_CONSUME_V1.feature_df
# Should give the renamed Features

FV_USER_CONSUME_V1.feature_df.queries
# Shows the query defined in the FeatureView renaming features, and referencing the raw FeatureView database object