orca-globalizer 0.1.0

Convert ORCA node-orbit vectors (15/73) into global graphlet count vectors.

orca-globalizer

Convert ORCA per-node orbit vectors into global graphlet count vectors.

Install (editable)

pip install -e .

Usage

import numpy as np from orca_globalizer import globalize_orca

node_vectors = np.loadtxt("orca_output.txt") # N x 15 (orca_node=4) or N x 73 (orca_node=5) g = globalize_orca(node_vectors, orca_node=5, method="sum", log_transform=False) print(g.shape) # (30,)

Output modes: method="sum" vs method="frequency"

ORCA produces per-node orbit counts: a matrix of shape (N nodes × O orbits) where each entry tells you how many times that node participates in that orbit.

This package converts those node-level orbit counts into a global graphlet vector.

method="sum" (global graphlet counts)

Returns global counts of each graphlet type (e.g., how many triangles, how many 4-cycles, etc.).

How it works:

1. Sum orbit counts across all nodes (total “node-incidences” per orbit).
2. Group orbit columns that belong to the same graphlet and sum them.
3. Divide by the number of nodes in that graphlet (2 / 3 / 4 / 5) to undo multi-counting.

Why the division? Each graphlet instance touches k nodes, so when you sum over nodes it contributes k times.
Example: 10 triangles → 30 node-incidences (3 per triangle) → 30 / 3 = 10 triangles.

method="frequency" (normalized distribution)

Returns a normalized version of the global counts so the vector sums to 1:

freq[i] = count[i] / sum(count)

Interpretation:

• freq[i] is the fraction of all counted graphlets (in the output vector) that are of type Gi.
• Useful for comparing graphs of different sizes/densities.

Log transform (optional)

If log_transform=True, the package applies a log1p transform (i.e., log(1 + x)) to reduce heavy-tailed dominance.

For method="frequency":

• If log_before_normalize=True (default): log then normalize
• If log_before_normalize=False: normalize then log