I have written a version of k-means with map/reduce logic that doesn’t seem to work as it should,it is not finished yet so don’t worry if it doesn’t look like k-means.
The reduce phaze is written with Thrust and works just like it should (just like the cpu version),but the map stage produces different results in GPU and CPU.
I’m attaching the code file for you to see and help me.
my_k-means_cuda.mapgpu.cu (10 KB)
Many thanks in advance.