Changes for manuscript update. #1
Conversation
… bump version to already match text; re-run notebooks;
| cluster_tree[cluster_tree['child'] == parent]['parent'], | ||
| deaths | ||
| ) | ||
| cpdef np.ndarray simplify_hierarchy(np.ndarray condensed_tree, |
There was a problem hiding this comment.
The persistence threshold for branches was incorrect. It was easier to write a hierarchy simplification function than to repair epsilon_search.
| return labels, branch_labels | ||
|
|
||
|
|
||
| def branch_membership_from_centrality(branch_centrality_vectors): | ||
| def branch_membership_from_centrality(branch_centrality_vectors, temperature=1.0): |
There was a problem hiding this comment.
Added temperature parameter to control softmax.
| return result | ||
|
|
||
|
|
||
| class BranchCondensedTree(_BaseCondensedTree): |
There was a problem hiding this comment.
Added own versions of the CondensedTree class that colour leaves by their labels.
| label_values = np.unique(clusterer.labels_[points]) | ||
| noise_points = clusterer.labels_[points] == label_values[-1] | ||
| branch_labels[points[noise_points]] = np.argmax(membership[noise_points], axis=1) | ||
| labels[points[noise_points]] = label_values[branch_labels[points[noise_points]]] - offset | ||
| offset += int(len(label_values) > len(persistences)) |
There was a problem hiding this comment.
Only update labels for noise points.
|
|
||
| # Extract unique edges that stay within the cluster | ||
| edges = np.unique(edges[edges[:, 0] > -1.0, :], axis=0) |
There was a problem hiding this comment.
Fixes mutual reachability values in the ApproximationGraph
| return 1 / depths |
There was a problem hiding this comment.
Makes membership centrality behave as in-cluster centrality.
| @@ -85,7 +85,7 @@ def _compute_branch_linkage_of_cluster( | |||
| points = space_tree.data.base[cluster_points] | |||
| centroid = np.average(points, weights=cluster_probabilities, axis=0) | |||
| centralities = metric_fun.pairwise(centroid[None], points)[0, :] | |||
| centralities = centralities.max() - centralities | |||
| centralities = 1 / centralities | |||
There was a problem hiding this comment.
Simplify algorithm by making in-cluster eccentricity be the distance to centroid.
| @@ -1,6 +1,6 @@ | |||
| [metadata] | |||
| name = pyflasc | |||
| version = 0.1.2 | |||
| version = 0.1.3 | |||
There was a problem hiding this comment.
Version mentioned in the manuscript.
This PR contains all changes made to the code during a major revision of the Arxiv paper and prepares for version 0.1.3.