bad default parameters!

[fawazsammani]

can i know why on earth does the stable release have by default the number of k-means iterations as infinite, and assumes the condition of breaking the loop will be met??? You caused my GPU to run k-means infinitely for a single sample without breaking, for the whole weekend, and wasted 2 days of running experiments!

Either publish reliable code or don't touch your keyboard

[XA23i]

I got the same problem, and it just got stuck at the while loop.

[XA23i]

have you fixed it yet?

[fawazsammani]

@XA23i I installed the dev:
pip install git+https://github.com/subhadarship/kmeans_pytorch.git

Then you can specify the iteration limits:
cluster_ids, cluster_centers = kmeans(X, num_clusters=K, distance='euclidean', device=device, tqdm_flag = False, iter_limit = 500)

However, I really do not recommend using his code, as its not reliable. You never know what other problems there are. I found out (too late unfortunately) that meta offers a very efficient library called faiss. You can use k-means very simply:

kmeans = faiss.Kmeans(X, k=K, niter=500, nredo=10)
kmeans.train(X.astype(np.float32))
distances, cluster_ids = kmeans.index.search(X, 1)

[XA23i]

Thank you, I'll try it.