Removing Dangerous Features

Big thanks, wigglemuse!

We’ve discussed before, the tight correlation between features x, x+210, … , x + 840; we can see that in play here:

194, 194+210 = 404, 194 + 420 = 614, 194 + 630 = 824, 194 + 840 = 1034;
209, 209 + 210 = 419, 209 + 420 = 629, 209 + 630 = 839, 209 + 840 = 1049.

So actually just two out 210 features, if one “unredundifies” v3. And there were no “dangerous” features features, from the 149 added in v4?

Yeah, the first 1050 features in v4 are more-or-less the same 1050 features in v3 in the same order but with different names. (MikeP tells us some of them are slightly different, but essentially they are the same set.)

Got home last week from holiday and had a chance to check whether or not those features are also in the small and medium feature-set. The answer is a yes. :slight_smile: And also 6 features in the small feature-set, so I guess maybe not a good choice to try that one for now…

I am just really curious what those features can be…

Car-brand driven by the CEO? Average response time at the helpdesk? Something in the data provider pipeline borked?

1 Like