Kind of baseline approach to just cluster the crash location and check if most of the accidents happen around 6 hotspot?
yes i tried it quickly with no tuning, and it did get me on to 50 score (in 5/10/2020), but normal grediant desent did better job and get me to 40 score
cool, same experience. Will try after adding in the other data points
how did you write the code for generating the sample submission. abit stuck on that.
you have to predict the six positions for each 3 hour window in the submission file. Just replace the coordinates in the provided sample submission file.
Yes, me too , got a result of 51. will try normal gradient descent now.
Hi! :) Was wondering, how do you use gradient descent to tackle this challenge?
Just the same question haha