What's the best lb score you achieved using only clustering and coordinates of crashes? Mine is 44.29.
Note that i don't even use the time component in my current solution meaning that the locations for the ambulances is static time-wise across the submission file.
Hello @blenz
What clustering algorithm did you use ?
KMeans
I tried Kmeans too but I am getting 105.81
Thinking of ways to improve the score
A default model using kmeans on the crash coordinates with 6 clusters gives around 52.xx. Make sure you're not making a mistake in the creation of the submission file or something
I will do that. Thank you
I have been using the offline scoring function to score my model. I just submitted my submission file to Zindi, and I got 49.xx
yes. with a little tweaking, you should get a lower score.
Hi Blenz, i got 40.42. I modified kmeans by using the geometric median instead means. Kmeans minimizes the squared distances, but the score function is without squares. In fact, it is a well known optimization problem, called the "Multi Source Weber Problem" (MWP).
Thanks for your answer. I've read a little about MWP.
@Marvin27, what library are you using to create your models ?
Hi Bilaerl, because i am a total beginner with python, i started with a pure java program.