File size: 10,012 Bytes
95fa396 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 | """Data generation and loading for experiments."""
import numpy as np
from typing import List, Tuple, Dict, Optional
from collections import defaultdict
from src.graph_utils import (
generate_bounded_degree_graph, generate_erdos_renyi_graph,
generate_power_law_graph, build_adjacency
)
def generate_gamma_poisson_data(N, M, K, graph_type, avg_degree,
count_scale, a0, b0, c0, d0,
seed=0, keep_zeros=False):
"""Generate synthetic Gamma-Poisson matrix factorization data."""
rng = np.random.RandomState(seed)
U_true = rng.gamma(a0, 1.0 / b0, size=(N, K))
V_true = rng.gamma(c0, 1.0 / d0, size=(M, K))
if graph_type == 'bounded_degree':
graph_edges = generate_bounded_degree_graph(N, M, avg_degree, seed)
elif graph_type == 'erdos_renyi':
graph_edges = generate_erdos_renyi_graph(N, M, avg_degree, seed)
elif graph_type == 'power_law':
graph_edges = generate_power_law_graph(N, M, avg_degree, seed)
else:
raise ValueError(f"Unknown graph type: {graph_type}")
edges = []
for i, j in graph_edges:
rate = count_scale * np.dot(U_true[i], V_true[j])
x = rng.poisson(max(rate, 1e-10))
if x > 0 or keep_zeros:
edges.append((i, j, int(x)))
return edges, U_true, V_true, graph_edges
def generate_gaussian_gaussian_data(N, M, K, graph_type, avg_degree,
sigma_U, sigma_V, sigma_x, seed=0):
"""Generate synthetic Gaussian-Gaussian MF data."""
rng = np.random.RandomState(seed)
U_true = rng.normal(0, sigma_U, size=(N, K))
V_true = rng.normal(0, sigma_V, size=(M, K))
if graph_type == 'bounded_degree':
graph_edges = generate_bounded_degree_graph(N, M, avg_degree, seed)
elif graph_type == 'erdos_renyi':
graph_edges = generate_erdos_renyi_graph(N, M, avg_degree, seed)
elif graph_type == 'power_law':
graph_edges = generate_power_law_graph(N, M, avg_degree, seed)
else:
raise ValueError(f"Unknown graph type: {graph_type}")
edges = []
for i, j in graph_edges:
mean = np.dot(U_true[i], V_true[j])
x = rng.normal(mean, sigma_x)
edges.append((i, j, float(x)))
return edges, U_true, V_true, graph_edges
def generate_gaussian_gamma_data(N, M, K, graph_type, avg_degree,
a0, b0, c0, d0, sigma_x, seed=0):
"""Generate synthetic Gaussian likelihood + Gamma prior data."""
rng = np.random.RandomState(seed)
U_true = rng.gamma(a0, 1.0 / b0, size=(N, K))
V_true = rng.gamma(c0, 1.0 / d0, size=(M, K))
if graph_type == 'bounded_degree':
graph_edges = generate_bounded_degree_graph(N, M, avg_degree, seed)
elif graph_type == 'erdos_renyi':
graph_edges = generate_erdos_renyi_graph(N, M, avg_degree, seed)
elif graph_type == 'power_law':
graph_edges = generate_power_law_graph(N, M, avg_degree, seed)
else:
raise ValueError(f"Unknown graph type: {graph_type}")
edges = []
for i, j in graph_edges:
mean = np.dot(U_true[i], V_true[j])
x = rng.normal(mean, sigma_x)
edges.append((i, j, float(x)))
return edges, U_true, V_true, graph_edges
def load_lastfm_data(max_users=2000, max_items=2000, max_edges=100000,
min_user_degree=5, min_item_degree=5, max_count=100, seed=42):
"""Load Last.fm user-artist counts from HF dataset."""
from datasets import load_dataset
print("Loading Last.fm dataset...")
ds = load_dataset("matthewfranglen/lastfm-1k", split="train")
user_artist_counts = defaultdict(lambda: defaultdict(int))
for row in ds:
uid = row['user_index']
aid = row['artist_index']
user_artist_counts[uid][aid] += 1
user_degrees = {u: len(v) for u, v in user_artist_counts.items()}
valid_users = [u for u, d in user_degrees.items() if d >= min_user_degree]
item_degree = defaultdict(int)
for u in valid_users:
for a in user_artist_counts[u]:
item_degree[a] += 1
valid_items = set(a for a, d in item_degree.items() if d >= min_item_degree)
rng = np.random.RandomState(seed)
valid_users = sorted(valid_users)
if len(valid_users) > max_users:
valid_users = list(rng.choice(valid_users, max_users, replace=False))
valid_users_set = set(valid_users)
all_items = set()
for u in valid_users:
for a in user_artist_counts[u]:
if a in valid_items:
all_items.add(a)
all_items = sorted(all_items)
if len(all_items) > max_items:
all_items = list(rng.choice(all_items, max_items, replace=False))
valid_items_set = set(all_items)
user_map = {u: idx for idx, u in enumerate(sorted(valid_users_set))}
item_map = {a: idx for idx, a in enumerate(sorted(valid_items_set))}
edges = []
for u in valid_users_set:
for a, count in user_artist_counts[u].items():
if a in valid_items_set:
c = min(count, max_count)
if c > 0:
edges.append((user_map[u], item_map[a], int(c)))
if len(edges) > max_edges:
indices = rng.choice(len(edges), max_edges, replace=False)
edges = [edges[i] for i in indices]
N = len(user_map)
M = len(item_map)
preprocessing = {
'dataset': 'matthewfranglen/lastfm-1k', 'N': N, 'M': M,
'n_edges': len(edges), 'max_count': max_count, 'seed': seed,
}
print(f"Last.fm loaded: N={N}, M={M}, edges={len(edges)}")
return edges, N, M, preprocessing
def load_movielens_data(mode='rating_count', max_users=2000, max_items=2000,
max_edges=100000, min_user_degree=5, min_item_degree=5, seed=42):
"""Load MovieLens ratings from HF dataset."""
from datasets import load_dataset
print("Loading MovieLens dataset...")
ds = load_dataset("ashraq/movielens_ratings", split="train")
rng = np.random.RandomState(seed)
user_item_ratings = defaultdict(dict)
for row in ds:
uid = row['user_id']
mid = row['movie_id']
rating = row['rating']
user_item_ratings[uid][mid] = rating
user_degrees = {u: len(v) for u, v in user_item_ratings.items()}
valid_users = [u for u, d in user_degrees.items() if d >= min_user_degree]
item_degree = defaultdict(int)
for u in valid_users:
for m in user_item_ratings[u]:
item_degree[m] += 1
valid_items = set(m for m, d in item_degree.items() if d >= min_item_degree)
if len(valid_users) > max_users:
valid_users = list(rng.choice(valid_users, max_users, replace=False))
valid_users_set = set(valid_users)
all_items = set()
for u in valid_users_set:
for m in user_item_ratings[u]:
if m in valid_items:
all_items.add(m)
all_items = sorted(all_items)
if len(all_items) > max_items:
all_items = list(rng.choice(all_items, max_items, replace=False))
valid_items_set = set(all_items)
user_map = {u: idx for idx, u in enumerate(sorted(valid_users_set))}
item_map = {m: idx for idx, m in enumerate(sorted(valid_items_set))}
edges = []
for u in valid_users_set:
for m, rating in user_item_ratings[u].items():
if m in valid_items_set:
if mode == 'rating_count':
x = int(np.ceil(rating))
elif mode == 'binary':
x = 1
else:
raise ValueError(f"Unknown mode: {mode}")
if x > 0:
edges.append((user_map[u], item_map[m], x))
if len(edges) > max_edges:
indices = rng.choice(len(edges), max_edges, replace=False)
edges = [edges[i] for i in indices]
N = len(user_map)
M = len(item_map)
preprocessing = {
'dataset': 'ashraq/movielens_ratings', 'mode': mode,
'N': N, 'M': M, 'n_edges': len(edges), 'seed': seed,
}
print(f"MovieLens ({mode}) loaded: N={N}, M={M}, edges={len(edges)}")
return edges, N, M, preprocessing
def sample_deletions(edges, user_to_items, item_to_users, num_deletions, seed=0):
"""Sample deletions with 25% each: random, high-count, hub-adjacent, low-degree."""
rng = np.random.RandomState(seed)
n_per_type = num_deletions // 4
remainder = num_deletions - 4 * n_per_type
counts = np.array([e[2] for e in edges], dtype=float)
user_degrees = defaultdict(int)
item_degrees = defaultdict(int)
for i, j, x in edges:
user_degrees[i] += 1
item_degrees[j] += 1
hub_scores = np.array([max(user_degrees[e[0]], item_degrees[e[1]]) for e in edges], dtype=float)
low_deg_scores = np.array([min(user_degrees[e[0]], item_degrees[e[1]]) for e in edges], dtype=float)
sampled = []
used = set()
def _sample(scores, n, dtype, high=True):
avail = [i for i in range(len(edges)) if i not in used]
if not avail or n <= 0:
return
sc = scores[avail]
if high:
ranked = np.argsort(-sc)
else:
ranked = np.argsort(sc)
pool = ranked[:min(len(avail), max(n * 3, 20))]
chosen = rng.choice(pool, size=min(n, len(pool)), replace=False)
for idx in chosen:
eidx = avail[idx]
used.add(eidx)
sampled.append((edges[eidx], dtype))
# Random
avail = list(range(len(edges)))
rng.shuffle(avail)
for idx in avail[:n_per_type + remainder]:
used.add(idx)
sampled.append((edges[idx], 'random'))
_sample(counts, n_per_type, 'high_count', high=True)
_sample(hub_scores, n_per_type, 'hub_adjacent', high=True)
_sample(low_deg_scores, n_per_type, 'low_degree', high=False)
return sampled
|