Jonahs dating dating a rules girl
But there were still (weak) well-defined group consensuses on a person’s relative ambition, intelligence and sincerity.
In this context, random forest models are unusually nontransparent on account of their ability to infer the identities of the events' participants, and unless one includes the same participant in both the train and test sets, they perform very poorly.
I gradually shifted toward focusing on understanding what the data tells us about human diversity.
Following this line of thought ultimately facilitated the creation of a better predictive model, while having relevance extending beyond the context of speed dating. I quantified the extent to which there was a universal consensus on a participant’s relative attractiveness, finding that 60% of the variation in perceptions of relative attractiveness can be attributed to the group consensus while 40% was accounted for by individual idiosyncrasies of raters.
This is considerably more involved than it would be if it wasn't necessary to exclude the two participants' ratings of each other.
In fact, simply excluding these ratings introduces structure in the dataset that results in the features being contaminated with the participants' ratings of each other in a subtle way, and we introduce a stochastic element to eliminate the contamination.