Test dispersion features

[dfsnow]

Predictive models with a lot of heterogeneity in the training data often include features that measure the dispersion/variation of each predictor. For example, in addition to a building's year built, you might also include the standard deviation of the year built, where the aggregation group is all properties within 1000 meters.

[Damonamajor]

@dfsnow Can you give a quick list on metrics you think this would be valuable for?

[dfsnow]

Basically any continuous feature where we think the variance of the feature will have some effect on local values. Start with char_bldg_sf and char_yrblt. I'd expect both of those to be impactful here, as you can say things like "The average year built for this area is high and the variance is low, indicating this neighborhood is almost entirely new construction (and therefore likely worth more)."