-
Notifications
You must be signed in to change notification settings - Fork 0
/
similarity_metric.txt
9 lines (9 loc) · 1.2 KB
/
similarity_metric.txt
1
2
3
4
5
6
7
8
9
The reasoning behind the chosen similarity metric.
In developing a similarity metric, I chose to use columns that were not sparsely populated.
If those columns were used when variables were not present, there would be days that similarity could not be calculated.
The name of the station and location information(longitude / latitude / elevation) do not offer much in explaining the variance for environmental conditions, so they were not considered for this metric.
Wind speed is represented by the day's windchill, this prevents the Euclidean distance measure from placing too much importance on wind conditions while accounting for what a human would perceive as a similar day.
Humidity is represented by the dew point temperature instead of the relative humidity because the dew point temperature is absolute, captures humidity in its definition, and is on a similar scale to most of the other variables selected for the similarity metric.
Altimeter Setting was used over station pressure because it reads at a mean sea level, unlike station pressure.
The formulation for the similarity index is currently
(WindchillTemperatureF^2 + DryBulbTemperatureF^2 + WetBulbTemperatureF^2 + DewPointTemperatureF^2 + AltimeterSetting^2)^.5