initial commit

codes/experiments/synthetic_data/graph_generators/base_graph_generator.py

@@ -0,0 +1,36 @@
+import numpy as np
+
+def generate_graph(NODES=10, DENSITY=0.1, DISTRIBUTION=('lognormal', (100, 1)), SEED=42):
+    rng = np.random.default_rng(SEED)
+
+    # Calculate variables
+    max_edges = (NODES - 1) * NODES
+    edges = int(round(DENSITY * max_edges))
+
+    # Make edges
+    sel = rng.choice(max_edges, size=edges, replace=False)
+
+    # Make distance array
+    dist_name, dist_params = DISTRIBUTION
+
+    if dist_name == "lognormal":
+        mean, sigma = dist_params
+        distances = rng.lognormal(mean=mean, sigma=sigma, size=edges)
+    elif dist_name == "uniform":
+        mean = dist_params[0]
+        distances = rng.uniform(0.0, 2.0 * mean, size=edges)
+    elif dist_name == "exponential":
+        mean = dist_params[0]
+        distances = rng.exponential(scale=mean, size=edges)
+    else:
+        raise ValueError("Unknown DISTRIBUTION")
+
+    # Make adj
+    adj = [[] for _ in range(NODES)]
+    for idx, dist in zip(sel, distances):
+        u = idx // (NODES-1)
+        r = idx % (NODES-1)
+        v = r if r < u else r + 1
+        adj[u].append((v, dist))
+
+    return adj

codes/experiments/synthetic_data/graph_generators/outdegree_graph_generator.py

@@ -0,0 +1,44 @@
+import numpy as np
+
+def outdegree_generate_graph(NODES, DENSITY, DISTRIBUTION, SEED=42):
+    rng = np.random.default_rng(SEED)
+
+    max_edges = (NODES - 1) * NODES
+    edges = int(round(DENSITY * max_edges))
+
+    base = edges // NODES
+    rem = edges % NODES
+
+    adj = [[] for _ in range(NODES)]
+
+    dist_name, dist_params = DISTRIBUTION
+
+    if dist_name == "lognormal":
+        mean, sigma = dist_params
+        dist_func = lambda size: rng.lognormal(mean=mean, sigma=sigma, size=size)
+    elif dist_name == "uniform":
+        mean = dist_params[0]
+        dist_func = lambda size: rng.uniform(0.0, 2.0 * mean, size=size)
+    elif dist_name == "exponential":
+        mean = dist_params[0]
+        dist_func = lambda size: rng.exponential(scale=mean, size=size)
+    else:
+        raise ValueError("Unknown DISTRIBUTION")
+
+    for u in range(NODES):
+        d_u = base + (1 if u < rem else 0)
+
+        if d_u == 0:
+            continue
+
+        targets = rng.choice(NODES-1, size=d_u, replace=False)
+        targets = np.where(targets < u, targets, targets + 1)
+
+        weights = dist_func(d_u)
+        weights = np.rint(weights).astype(np.int64)
+        weights = np.maximum(weights, 1)
+
+        for v, w in zip(targets, weights):
+            adj[u].append((v, float(w)))
+
+    return adj