droyed / eucl_dist Goto Github PK

Sometimes a fast cosine similarity instead of euclidean distance is also useful. It would be very handy to have such an implementation.

I want to know how did the author deal with Numerical stability problem? as mentioned by https://stackoverflow.com/questions/34110493/is-scipy-cdist-efficient# :
"Try this yourself on a Python interpreter: a = 1e15 +1; b = 1e15, now compute (a - b)2 and a2 + b**2 - 2ab. Surprised? You have just witnessed (lack of) numerical stability at work. So unless you are certain this type of problem cannot happen in your dataset, it is better to sacrifice performance for correctness. "– Jaime Dec 6 '15 at 16:59

Hi,
I have cuda11.1 installed on windows 10
and i have installed pycuda-2020.1-py3.6

this error showed for me.
Any suggestions?

Traceback (most recent call last):
File "C:/Users/s-f2/Desktop/DELF/gpu_euclidean/main.py", line 4, in
from eucl_dist.gpu_dist import dist as gdist
File "C:\Users\s-f2\Desktop\DELF\gpu_euclidean\eucl_dist\gpu_dist.py", line 1, in
import pycuda.gpuarray as gpuarray
File "C:\Users\s-f2\Desktop\DELF\gpu_euclidean\venv\lib\site-packages\pycuda\gpuarray.py", line 4, in
import pycuda.elementwise as elementwise
File "C:\Users\s-f2\Desktop\DELF\gpu_euclidean\venv\lib\site-packages\pycuda\elementwise.py", line 35, in
from pycuda.tools import context_dependent_memoize
File "C:\Users\s-f2\Desktop\DELF\gpu_euclidean\venv\lib\site-packages\pycuda\tools.py", line 34, in
import pycuda.driver as cuda
File "C:\Users\s-f2\Desktop\DELF\gpu_euclidean\venv\lib\site-packages\pycuda\driver.py", line 62, in
from pycuda._driver import * # noqa
ImportError: DLL load failed: The specified procedure could not be found.

Process finished with exit code 1

Hello @droyed
I have face encodings 512 points for one face I want to calculate euclidean distance of them using this code any idea?