File size: 19,485 Bytes
bf620c6 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 |
package old;
import java.io.BufferedReader;
import java.io.BufferedWriter;
import java.io.IOException;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.*;
/**
* Cora ablation driver for L-RMC.
* <p>
* - Loads Cora (content + cites)
* - Calls clique2_ablations.runLaplacianRMC(adj) to get reconstruction snapshots
* - Scores snapshots with the calibrated surrogate:
* \tilde S_L(C) = |C| * ( dbar(C) - alpha * sqrt(Q + epsilon) )
* where Q = d^T L_C d from snapshots, dbar = sumDegIn / |C|
* - For each (epsilon, alpha) combination, assigns each node the majority label
* of the highest-scoring snapshot that contains it, then reports accuracy.
* <p>
* α choices match the ablation in the paper:
* α ∈ { diam(C), 1/√λ2(C) } and ε ∈ {1e-8, 1e-6, 1e-4}.
* <p>
* References to definitions and ablation settings:
* - Surrogate + α calibration and bounds: Section 5, eqs. (6)-(7), SeL.
* - Ablation settings on ε and α on Cora: Section 7.3 and Figure 3.
*/
class LRMCablations2 {
// Ablation grid
static final double[] EPSILONS = {1e-8, 1e-6, 1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4, 20000, 1e5, 1e6, 1e7, 1e8, 1e9, 1e10, 1e12, 1e13, 1e14};
public static void main(String[] args) throws Exception {
if (args.length < 3) {
System.err.println("Usage: java LRMCablations2 <path/to/cora.content> <path/to/cora.cites> <output_csv> <output_seeds> <alpha_type> <eps>");
return;
}
final Path contentPath = Paths.get(args[0]);
final Path citesPath = Paths.get(args[1]);
final Path outCsv = Paths.get(args[2]);
final Path outSeeds = (args.length >= 4 ? Paths.get(args[3]) : null);
final AlphaKind alphaKind = (args.length >= 5 ? parseAlpha(args[4]) : AlphaKind.DIAM);
final double eps = (args.length >= 6 ? Double.parseDouble(args[5]) : 1e-6);
// 1) Load Cora graph and labels
GraphData G = loadCora(contentPath, citesPath);
System.out.printf(Locale.US, "# Loaded Cora: n=%d, m=%d, classes=%d%n",
G.n, G.m, G.labelNames.length);
// 2) Run L-RMC reconstruction to get snapshots
List<clique2_ablations.SnapshotDTO> snaps = clique2_ablations.runLaplacianRMC(G.adj1Based);
System.out.printf(Locale.US, "# Reconstruction snapshots: %d%n", snaps.size());
// 3) Either sweep for the paper CSV, or compute a single row for export mode
List<ResultRow> outRows = new ArrayList<>();
if (outSeeds != null) {
// Single point using CLI epsilon/alpha
double acc = labelAccuracyFromSnapshots(snaps, G, eps, alphaKind);
String alphaName = (alphaKind == AlphaKind.DIAM) ? "diam(C)" : "1/sqrt(lambda2)";
outRows.add(new ResultRow(eps, alphaName, acc, G.n, G.m));
System.out.printf(Locale.US, "epsilon=%.0e, alpha=%s, accuracy=%.4f%n",
eps, alphaName, acc);
} else {
// Full sweep for ablations
for (double epsGrid : EPSILONS) {
for (AlphaKind aKind : AlphaKind.values()) {
double acc = labelAccuracyFromSnapshots(snaps, G, epsGrid, aKind);
String alphaName = (aKind == AlphaKind.DIAM) ? "diam(C)" : "1/sqrt(lambda2)";
outRows.add(new ResultRow(epsGrid, alphaName, acc, G.n, G.m));
System.out.printf(Locale.US, "epsilon=%.0e, alpha=%s, accuracy=%.4f%n",
epsGrid, alphaName, acc);
}
}
}
// 4) Save CSV
try (BufferedWriter w = Files.newBufferedWriter(outCsv, StandardCharsets.UTF_8)) {
w.write("epsilon,alpha,accuracy,nodes,edges\n");
for (ResultRow r : outRows) {
w.write(String.format(Locale.US, "%.0e,%s,%.6f,%d,%d%n",
r.epsilon, r.alpha, r.accuracy, r.n, r.m));
}
}
System.out.println("# Wrote: " + outCsv.toAbsolutePath());
if (outSeeds != null) {
exportLrmcSeeds(snaps, G, eps, alphaKind, outSeeds);
}
}
// Accuracy from α-calibrated surrogate over all nodes
static double labelAccuracyFromSnapshots(
List<clique2_ablations.SnapshotDTO> snaps,
GraphData G,
double epsilon,
AlphaKind alphaKind
) {
final int n = G.n;
final int numClasses = G.labelNames.length;
double[] bestScore = new double[n];
int[] bestLabel = new int[n];
Arrays.fill(bestScore, Double.NEGATIVE_INFINITY);
Arrays.fill(bestLabel, -1);
// Pre-allocate scratch to avoid churn in inner loops
boolean[] inC = new boolean[n];
for (clique2_ablations.SnapshotDTO s : snaps) {
final int[] nodes = s.nodes; // 0-based ids
final int k = nodes.length;
if (k == 0) continue;
// Mark current component
for (int u : nodes) inC[u] = true;
// Average internal degree in the component
final double dbar = (k == 0) ? 0.0 : (s.sumDegIn / (double) k);
// Laplacian energy Q and calibrated α
final double Q = s.Q;
double alpha;
if (alphaKind == AlphaKind.DIAM) {
alpha = approxDiameter(nodes, G.adj1Based, inC);
} else {
double lam2 = approxLambda2(nodes, G.adj1Based, inC);
if (lam2 <= 1e-12) lam2 = 1e-12; // guard for nC=1 or numerical zeros
alpha = 1.0 / Math.sqrt(lam2);
}
// Calibrated surrogate \tilde S_L(C)
final double score = k * (dbar - alpha * Math.sqrt(Q + epsilon));
// Majority label of this component
int maj = majorityLabel(nodes, G.labels, numClasses);
// Update node-wise best snapshot under this scoring
for (int u : nodes) {
if (score > bestScore[u]) {
bestScore[u] = score;
bestLabel[u] = maj;
}
}
// Unmark
for (int u : nodes) inC[u] = false;
}
// Compute accuracy over all nodes
int correct = 0;
for (int u = 0; u < n; u++) {
// Every node appears in at least its singleton snapshot
if (bestLabel[u] == G.labels[u]) correct++;
}
return correct / (double) n;
}
// Majority label
static int majorityLabel(int[] nodes, int[] labels, int numClasses) {
int[] cnt = new int[numClasses];
for (int u : nodes) cnt[labels[u]]++;
int best = 0, arg = 0;
for (int c = 0; c < numClasses; c++) {
if (cnt[c] > best) {
best = cnt[c];
arg = c;
}
}
return arg;
}
// Approximate diameter by 2-sweep BFS on the induced subgraph
static int approxDiameter(int[] nodes, List<Integer>[] adj1, boolean[] inC) {
if (nodes.length <= 1) return 0;
int start = nodes[0];
int u = farthestInComponent(start, adj1, inC).node;
BFSResult r = farthestInComponent(u, adj1, inC);
return r.dist;
}
static BFSResult farthestInComponent(int src0, List<Integer>[] adj1, boolean[] inC) {
int n = inC.length;
int[] dist = new int[n];
Arrays.fill(dist, -1);
ArrayDeque<Integer> q = new ArrayDeque<>();
dist[src0] = 0;
q.add(src0);
int bestNode = src0, bestDist = 0;
while (!q.isEmpty()) {
int u = q.poll();
int du = dist[u];
if (du > bestDist) {
bestDist = du;
bestNode = u;
}
for (int v1 : adj1[u + 1]) {
int v = v1 - 1;
if (!inC[v]) continue;
if (dist[v] >= 0) continue;
dist[v] = du + 1;
q.add(v);
}
}
return new BFSResult(bestNode, bestDist);
}
// Approximate λ2(L_C) via orthogonalized power iteration
static double approxLambda2(int[] nodes, List<Integer>[] adj1, boolean[] inC) {
final int k = nodes.length;
if (k <= 1) return 0.0;
// local index mapping
int[] loc = new int[inC.length];
Arrays.fill(loc, -1);
for (int i = 0; i < k; i++) loc[nodes[i]] = i;
// Build degrees inside C and find d_max
int[] deg = new int[k];
int dmax = 0;
for (int i = 0; i < k; i++) {
int u = nodes[i];
int du = 0;
for (int w1 : adj1[u + 1]) {
int w = w1 - 1;
if (loc[w] >= 0) du++;
}
deg[i] = du;
if (du > dmax) dmax = du;
}
if (dmax == 0) return 0.0;
final double c = 2.0 * dmax + 1.0; // shift
double[] x = new double[k];
Random rng = new Random(42);
for (int i = 0; i < k; i++) x[i] = rng.nextDouble() - 0.5;
orthToOnes(x);
normalize(x);
double[] Lx = new double[k];
double[] y = new double[k];
final int iters = 30;
for (int it = 0; it < iters; it++) {
// L x
Arrays.fill(Lx, 0.0);
for (int i = 0; i < k; i++) {
double sumNbr = 0.0;
int u = nodes[i];
for (int w1 : adj1[u + 1]) {
int j = loc[w1 - 1];
if (j >= 0) sumNbr += x[j];
}
Lx[i] = deg[i] * x[i] - sumNbr;
}
// y = x - (1/c)*Lx
for (int i = 0; i < k; i++) y[i] = x[i] - Lx[i] / c;
orthToOnes(y);
normalize(y);
System.arraycopy(y, 0, x, 0, k);
}
// Rayleigh quotient x^T L x
double num = 0.0;
for (int i = 0; i < k; i++) {
double sumNbr = 0.0;
int u = nodes[i];
for (int w1 : adj1[u + 1]) {
int j = loc[w1 - 1];
if (j >= 0) sumNbr += x[j];
}
double Lxi = deg[i] * x[i] - sumNbr;
num += x[i] * Lxi;
}
// x is unit norm, so denom ~ 1
return Math.max(num, 0.0);
}
static void orthToOnes(double[] v) {
double mean = 0.0;
for (double x : v) mean += x;
mean /= v.length;
for (int i = 0; i < v.length; i++) v[i] -= mean;
}
static void normalize(double[] v) {
double nrm2 = 0.0;
for (double x : v) nrm2 += x * x;
if (nrm2 <= 0) {
v[0] = 1.0;
nrm2 = 1.0;
}
double inv = 1.0 / Math.sqrt(nrm2);
for (int i = 0; i < v.length; i++) v[i] *= inv;
}
// Cora loader
static GraphData loadCora(Path content, Path cites) throws IOException {
Map<String, Integer> id2idx = new LinkedHashMap<>();
Map<String, Integer> lbl2idx = new LinkedHashMap<>();
List<String> lblNames = new ArrayList<>();
List<Integer> labelsList = new ArrayList<>();
// Pass 1: content defines node universe and labels
try (BufferedReader br = Files.newBufferedReader(content, StandardCharsets.UTF_8)) {
String s;
while ((s = br.readLine()) != null) {
s = s.trim();
if (s.isEmpty()) continue;
String[] tok = s.split("\\s+");
String id = tok[0];
String lab = tok[tok.length - 1];
int u = id2idx.computeIfAbsent(id, _k -> id2idx.size());
int c = lbl2idx.computeIfAbsent(lab, _k -> {
lblNames.add(lab);
return lblNames.size() - 1;
});
// Extend labels list to position u if needed
while (labelsList.size() <= u) labelsList.add(0);
labelsList.set(u, c);
}
}
int n = id2idx.size();
int[] labels = new int[n];
for (int i = 0; i < n; i++) labels[i] = labelsList.get(i);
// Temp adjacency as sets to dedup
@SuppressWarnings("unchecked")
HashSet<Integer>[] adjSet1 = new HashSet[n + 1];
for (int i = 1; i <= n; i++) adjSet1[i] = new HashSet<>();
// Pass 2: cites edges
long mUndir = 0;
try (BufferedReader br = Files.newBufferedReader(cites, StandardCharsets.UTF_8)) {
String s;
while ((s = br.readLine()) != null) {
s = s.trim();
if (s.isEmpty() || s.startsWith("#")) continue;
String[] tok = s.split("\\s+|,");
if (tok.length < 2) continue;
Integer ui = id2idx.get(tok[0]);
Integer vi = id2idx.get(tok[1]);
if (ui == null || vi == null) continue; // skip unknown ids
int a = ui + 1, b = vi + 1; // to 1-based
if (a == b) continue;
if (adjSet1[a].add(b)) {
adjSet1[b].add(a);
mUndir++;
}
}
}
@SuppressWarnings("unchecked")
List<Integer>[] adj1 = new ArrayList[n + 1];
for (int i = 1; i <= n; i++) {
adj1[i] = new ArrayList<>(adjSet1[i]);
}
GraphData G = new GraphData();
G.n = n;
G.m = mUndir;
G.adj1Based = adj1;
G.labels = labels;
G.labelNames = lblNames.toArray(new String[0]);
return G;
}
// Parse alpha kind from string
static AlphaKind parseAlpha(String s) {
String t = s.trim().toUpperCase(Locale.ROOT);
if (t.equals("DIAM") || t.equals("DIAM(C)") || t.equals("DIAMETER")) return AlphaKind.DIAM;
if (t.equals("INV_SQRT_LAMBDA2") || t.equals("1/SQRT(LAMBDA2)") || t.equals("LAMBDA2"))
return AlphaKind.INV_SQRT_LAMBDA2;
throw new IllegalArgumentException("Unknown alpha kind: " + s);
}
// Export seeds as a node->cluster partition built from best snapshot per node
static void exportLrmcSeeds(
List<clique2_ablations.SnapshotDTO> snaps,
GraphData G,
double epsilon,
AlphaKind alphaKind,
Path outJson) throws IOException {
final int n = G.n;
final boolean[] inC = new boolean[n];
// 1) Score every snapshot with the calibrated surrogate: k*(dbar - alpha*sqrt(Q+eps))
final double[] snapScore = new double[snaps.size()];
Arrays.fill(snapScore, Double.NEGATIVE_INFINITY);
for (int i = 0; i < snaps.size(); i++) {
clique2_ablations.SnapshotDTO s = snaps.get(i);
final int[] nodes = s.nodes; // 0-based ids
final int k = nodes.length;
if (k == 0) continue;
for (int u : nodes) inC[u] = true;
final double dbar = s.sumDegIn / Math.max(1.0, k);
final double Q = s.Q;
final double alpha;
if (alphaKind == AlphaKind.DIAM) {
alpha = approxDiameter(nodes, G.adj1Based, inC);
} else {
double lam2 = approxLambda2(nodes, G.adj1Based, inC);
if (lam2 <= 1e-12) lam2 = 1e-12;
alpha = 1.0 / Math.sqrt(lam2);
}
final double score = k * (dbar - alpha * Math.sqrt(Q + epsilon));
snapScore[i] = score;
for (int u : nodes) inC[u] = false;
}
// 2) Best snapshot per node
final double[] best = new double[n];
final int[] bestSnap = new int[n];
Arrays.fill(best, Double.NEGATIVE_INFINITY);
Arrays.fill(bestSnap, -1);
for (int i = 0; i < snaps.size(); i++) {
clique2_ablations.SnapshotDTO s = snaps.get(i);
double sc = snapScore[i];
for (int u : s.nodes) {
if (sc > best[u]) {
best[u] = sc;
bestSnap[u] = i;
}
}
}
// 3) Build cluster membership map: snapshot id -> members
LinkedHashMap<Integer, List<Integer>> members = new LinkedHashMap<>();
for (int u = 0; u < n; u++) {
int sid = bestSnap[u];
if (sid < 0) continue; // should not happen if singleton snapshots exist
members.computeIfAbsent(sid, _k -> new ArrayList<>()).add(u);
}
// 4) Relabel to contiguous cluster ids
LinkedHashMap<Integer, Integer> snap2cluster = new LinkedHashMap<>();
int cid = 0;
for (int sid : members.keySet()) snap2cluster.put(sid, cid++);
// 5) Write JSON
try (BufferedWriter w = Files.newBufferedWriter(outJson, StandardCharsets.UTF_8)) {
w.write("{\"meta\":{");
w.write("\"epsilon\":" + epsilon + ",\"alpha_kind\":\"" + alphaKind + "\",\"n\":" + G.n + ",\"m\":" + G.m + "},");
w.write("\"clusters\":[\n");
boolean first = true;
for (Map.Entry<Integer, List<Integer>> e : members.entrySet()) {
int sid = e.getKey();
int clusterId = snap2cluster.get(sid);
clique2_ablations.SnapshotDTO s = snaps.get(sid);
if (!first) w.write(",\n");
first = false;
w.write(" {\"cluster_id\":" + clusterId);
w.write(",\"snapshot_id\":" + sid);
w.write(",\"score\":" + snapScore[sid]);
w.write(",\"k_seed\":" + s.nodes.length);
w.write(",\"members\":" + intListToJson(e.getValue()));
w.write(",\"seed_nodes\":" + intArrayToJson(s.nodes));
w.write("}");
}
w.write("\n]}");
}
System.out.println("# Wrote seeds: " + outJson.toAbsolutePath());
}
static String intArrayToJson(int[] a) {
StringBuilder sb = new StringBuilder();
sb.append('[');
for (int i = 0; i < a.length; i++) {
if (i > 0) sb.append(',');
sb.append(a[i]);
}
sb.append(']');
return sb.toString();
}
static String intListToJson(List<Integer> a) {
StringBuilder sb = new StringBuilder();
sb.append('[');
for (int i = 0; i < a.size(); i++) {
if (i > 0) sb.append(',');
sb.append(a.get(i));
}
sb.append(']');
return sb.toString();
}
enum AlphaKind {DIAM, INV_SQRT_LAMBDA2}
static class BFSResult {
final int node, dist;
BFSResult(int node, int dist) {
this.node = node;
this.dist = dist;
}
}
// Data holders
static final class GraphData {
int n;
long m;
List<Integer>[] adj1Based; // 1-based adjacency
int[] labels; // 0-based labels per node id
String[] labelNames;
}
static final class ResultRow {
final double epsilon;
final String alpha;
final double accuracy;
final int n;
final long m;
ResultRow(double epsilon, String alpha, double accuracy, int n, long m) {
this.epsilon = epsilon;
this.alpha = alpha;
this.accuracy = accuracy;
this.n = n;
this.m = m;
}
}
}
|