mlandry22 / av-hikeathon Goto Github PK

The following screenshot shows the results of four different Driverless AI experiments.

Where train4 uses MOD_SPLIT<-0 in

And the accuracy settings of 5 and 6 were used, with the impact being that all 6 million rows were used for the 6/2/1 models and about 80% or so used for the 5/2/1, and also the ensembling step is more advanced.

The models are shown in more detail in the four comments below.

user1 chat rate as reciprocal. so let's again take the "rohan,mark,?". Just getting the value for "mark.rohan,[target]" as a feature is useful. could also be useful to dig one more step in and see how well node1 usually does when in reciprocal relationships. So if "rohan" usually responds when people send him a chat, that's useful to know.