feat: add string_json value type for

[zerafachris]

What this PR does / why we need it:

On-demand feature views are very powerful, but we can extend their functionality further by allowing a string of JSON as a type.

Consider a feature with Field(name='value_array', dtype=String_JSON).
An example feature would be {'0': 127.5433, '1': 183.2108, '2': 157.525, '3': 170.2931, '4': 186.0841, '5': 141.6574, '6': 146.1895, '7': 148.6707, '8': 146.0211, '9': 196.4125, '10': 132.7219, '11': 154.1899, '12': 132.3854}

The on-demand feature view now can be defined as:

@on_demand_feature_view(
    sources=[driver_hourly_stats_view, input_request],
    schema=[
        Field(name="value_array_modified", dtype=Float64)
    ],
)
def array_modified(inputs: pd.DataFrame) -> pd.DataFrame:
    df = pd.DataFrame()
    df['value_array_modified'] = 0.0
    for index, row in inputs.iterrows():
        x1 = eval(row['value_array']).get('1')
        x1 = x1 if x1 is not None else 0.0
        x2 = eval(row['value_array']).get('2')
        x2 = x2 if x2 is not None else 0.0
        x3 = eval(row['value_array']).get('3')
        x3 = x3 if x3 is not None else 0.0
        df.at[index, 'value_array_modified'] = x1 + x2 + x3

    return df

Upon requested this feature, the response would be 511.0289

Which issue(s) this PR fixes:

Misc

[franciscojavierarceo]

Nice! Would you mind adding a test to highlight the behavior? There's an on demand transformation test file.

[zerafachris]

Not sure how to fix the error "Error: The template is not valid. .github/workflows/pr_integration_tests.yml (Line: 91, Col: 16): hashFiles('**/py3.11-ci-requirements.txt') failed. Fail to hash files under directory '/home/runner/work/feast/feast'".

Any advise?

[franciscojavierarceo]

Actually that's not from you. I saw this on your PR and created a dummy PR to confirm this is happening to others.

I'm looking into it but won't block this PR.

[redhatHameed]

@zerafachris Can you fix the linter and unit test issues? Please refer to the comments within the test itself.

[zerafachris]

I am cancelling this PR as the value type is not needed. Doing pre-processing such as if ':' not in row["feature1"] would allow for the feature projection of random_input to be successful.