Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Improve memory management in clustering_qr.kmeans_plusplus #775

Open
wants to merge 7 commits into
base: main
Choose a base branch
from
Open

Improve memory management in clustering_qr.kmeans_plusplus #775

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
5169a39
Improve memory management in clustering_qr.kmeans_plusplus
RobertoDF Sep 4, 2024
73afacf
Merge branch 'MouseLand:main' into Improved_memory_management_cluster…
RobertoDF Sep 6, 2024
73591f3
clear_cache arg in kmeans_plusplus
RobertoDF Sep 6, 2024
c7711ca
clear_cache arg in clustering_qr.run
RobertoDF Sep 6, 2024
0747ee8
Move tensor to CPU temporarily in kmeans_plusplus
RobertoDF Sep 6, 2024
d4bff08
Xd transfer to GPU moved within kmeans_plusplus, remove clear_cachhe arg
RobertoDF Sep 7, 2024
7680278
Refer to Xd not Xg ,Xg has been moved inside kmeans_plusplus
RobertoDF Sep 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 12 additions & 8 deletions kilosort/clustering_qr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,20 +127,18 @@ def cluster(Xd, iclust = None, kn = None, nskip = 20, n_neigh = 10, nclust = 200

m, ki, kj = Mstats(M, device=device)

#Xg = torch.from_numpy(Xd).to(dev)
Xg = Xd.to(device)
kn = torch.from_numpy(kn).to(device)

n_neigh = kn.shape[1]
NN, nfeat = Xg.shape
NN, nfeat = Xd.shape
nsub = (NN-1)//nskip + 1

rows_neigh = torch.arange(NN, device = device).unsqueeze(-1).tile((1,n_neigh))

tones2 = torch.ones((NN, n_neigh), device = device)

if iclust is None:
iclust_init = kmeans_plusplus(Xg, niter = nclust, seed = seed, device=device)
iclust_init = kmeans_plusplus(Xd, niter = nclust, seed = seed, device=device)
iclust = iclust_init.clone()
else:
iclust_init = iclust.clone()
Expand All @@ -162,9 +160,12 @@ def cluster(Xd, iclust = None, kn = None, nskip = 20, n_neigh = 10, nclust = 200
return iclust, isub, M, iclust_init


def kmeans_plusplus(Xg, niter = 200, seed = 1, device=torch.device('cuda')):
#Xg = torch.from_numpy(Xd).to(dev)
vtot = (Xg**2).sum(1)
def kmeans_plusplus(Xd, niter = 200, seed = 1, device=torch.device('cuda')):

Xg = Xd.to(device)

Xd_squared = (Xd ** 2)
vtot = Xd_squared.sum(1).to(device)

n1 = vtot.shape[0]
if n1 > 2**24:
Expand Down Expand Up @@ -199,7 +200,10 @@ def kmeans_plusplus(Xg, niter = 200, seed = 1, device=torch.device('cuda')):
isamp = torch.multinomial(v2, ntry)

Xc = Xg[isamp]
vexp = 2 * Xg @ Xc.T - (Xc**2).sum(1)
Xc_squared_sum = (Xc ** 2).sum(1)
vexp = Xg @ Xc.T
vexp.mul_(2)
vexp = vexp - Xc_squared_sum

dexp = vexp - vexp0.unsqueeze(1)
dexp = torch.relu(dexp)
Expand Down