Add clustering #68
Add clustering #68
Conversation
|
|
||
|
|
||
| def train_kmeans(features, n_clusters, path_to_kmeans, balanced=False, device='cpu'): | ||
| kmeans = BalancedKMeans(n_clusters=n_clusters, device=device, balanced=balanced) |
There was a problem hiding this comment.
want to note that on my local, n_clusters is not a parameter, but num_clusters is -- does this work for you?
There was a problem hiding this comment.
My bad - I did some refactoring on Huu's codebase and changed its name, will update it ASAP
There was a problem hiding this comment.
additionally, since I'm not using KMeans, but kmeans -- cannot be instantiated without X, which needs to be a torch object or else it fails
if this worked for you, great -- will sync up with you to figure out why it didn't on my env
if not, then we'll need to figure out a way to use pytorch-kmeans in batches since instantiation isn't immediately possible -- open to suggestions here
|
|
||
| from torch.utils.data import DataLoader, IterableDataset | ||
|
|
||
| from kmeans_pytorch import KMeans as BalancedKMeans |
There was a problem hiding this comment.
had issues with importing KMeans, turned up an error
but no issues importing kmeans -- did you experience this?
There was a problem hiding this comment.
Did you use the kmeans from pip or the one from https://github.com/kernelmachine/cbtm?
I've added some clustering experiments, be aware that to run the traditional clustering you need to install the kmeans from https://github.com/kernelmachine/cbtm