Trim un-helpful variables (or reduce their influence somehow?)

[blulightspecial]

See the latest version of the km_EDA.ipynb. A lot of the features that we are provided don't look like they're doing a whole lot. (not great variation or they're matching too closely with an existing variable)

• multicolinearity
  • h_road and h_fire are basically the same.
  • shade variables are very similar (and none have a lot of info).
  • ranked in terms of useful signal 3p > 9a > 12 p
• General lack of signal
  • vertical distance to hydrology looks equally distributed between all categories. It's not telling us anything
  • aspect
  • Generally, only seems like it's helpful with groups 4 and 6
  • degrees aspect had more information. If you look at the n/s and e/w you can't see any signal
  • The shade variables have generally weak signal.

[blulightspecial]

Most helpful variable by a mile is elevation (and some types of soil)