Skip to content

Instantly share code, notes, and snippets.

@weilinear
Created September 15, 2012 14:36
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save weilinear/3728246 to your computer and use it in GitHub Desktop.
Save weilinear/3728246 to your computer and use it in GitHub Desktop.
centers using sparse matrix
def _centers(X, labels, n_clusters, distances):
"""M step of the K-means EM algorithm
Computation of cluster centers / means.
Parameters
----------
X: array, shape (n_samples, n_features)
labels: array of integers, shape (n_samples)
Current label assignment
n_clusters: int
Number of desired clusters
Returns
-------
centers: array, shape (n_clusters, n_features)
The resulting centers
"""
# TODO: add support for CSR input
n_features = X.shape[1]
# TODO: explicit dtype handling
empty_cluster = {};
sparse_X = sp.issparse(X)
# label clone
reallocated_idx = 0
# deal with empty centers
far_from_centers = None
for center_id in range(n_clusters):
center_mask = labels == center_id
if sparse_X:
center_mask = np.arange(len(labels))[center_mask]
if not np.any(center_mask):
# Reassign empty cluster center to sample far from any cluster
if far_from_centers is None:
far_from_centers = distances.argsort()[::-1]
empty_cluster[center_id] = far_from_centers[reallocated_idx]
reallocated_idx += 1
# calculate indicator matrix
labels_unique, labels_normalized = np.unique(labels, return_inverse=True)
cluster_indicator = \
sp.csr_matrix((np.ones(labels.shape[0] + len(empty_cluster)),
np.array([np.concatenate([labels,
np.array(empty_cluster.keys())]),
np.concatenate([np.arange(X.shape[0]),
np.array(empty_cluster.values())])])),
shape=(n_clusters, X.shape[0]),
dtype=np.float)
# normalize cluster_indicator
inplace_csr_row_normalize_l1(cluster_indicator)
# cluster_indicator /= cluster_indicator.sum(axis=1)
centers = safe_sparse_dot(cluster_indicator, X, dense_output=True)
return centers
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment