search

src-d / kmcuda

Large scale K-means and K-nn implementation on NVIDIA GPU / CUDA
Other
783 stars 144 forks source link

Fail to run with python api #104

Closed max-ng closed 4 years ago

max-ng commented 4 years ago

kmeans = centroid, assignments, avg_distance = kmeans_cuda( database[:100], 4, yinyang_t=0 , verbosity=1, seed=4 ,device=1, average_distance=True)

log: reassignments threshold: 1 warning: p2p 1 <-> 0 is impossible warning: failed to enable p2p on gpu #0: peer access is not supported between these two devices warning: failed to enable p2p on gpu #1: peer access is not supported between these two devices transposing the samples... performing kmeans++... step 1 internal bug inside kmeans_init_centroids: dist_sum is NaN

max-ng commented 4 years ago

using one gpu, the log is now:

reassignments threshold: 1 transposing the samples... performing kmeans++... step 1 internal bug inside kmeans_init_centroids: dist_sum is NaN

internal bug in kmeans_init_centroids: j = 0 step 2/tmp/pip-68l6411e-build/src/kmeans.cu:814 -> an illegal memory access was encountered