File size: 15,121 Bytes
d792ef6
b6c4e22
 
09f9857
d792ef6
09f9857
d792ef6
09f9857
 
 
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
09f9857
d792ef6
09f9857
d792ef6
09f9857
d792ef6
09f9857
d792ef6
 
09f9857
d792ef6
09f9857
d792ef6
09f9857
d792ef6
 
09f9857
d792ef6
09f9857
d792ef6
 
 
 
09f9857
 
 
d792ef6
09f9857
d792ef6
 
 
09f9857
 
 
 
 
 
 
d792ef6
 
09f9857
 
 
 
d792ef6
09f9857
d792ef6
09f9857
 
d792ef6
09f9857
d792ef6
09f9857
 
d792ef6
 
09f9857
 
 
 
d792ef6
 
09f9857
d792ef6
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
09f9857
 
 
d792ef6
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
09f9857
d792ef6
 
 
 
09f9857
 
 
 
d792ef6
 
 
 
09f9857
 
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
 
d792ef6
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
09f9857
d792ef6
 
09f9857
d792ef6
 
 
 
09f9857
d792ef6
 
09f9857
 
 
d792ef6
 
 
 
09f9857
d792ef6
09f9857
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
09f9857
d792ef6
09f9857
d792ef6
 
 
 
 
09f9857
 
d792ef6
 
 
 
 
 
 
 
 
 
 
 
09f9857
 
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
 
 
 
09f9857
d792ef6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Fault; corruption from KL_Divergence corrupts the SVD and eigh solidity

SVAE - V=1024, D=24 (Validated Binding Constant)
==================================================
V=1024, D=24 -> CV=0.2916 (from sweep, confirmed)

Deep encoder/decoder for 1024x24 = 24,576 matrix.
Light KL on spectral shape (don't constrain magnitude).
Row CV should be ~0.29 by dimensional law.

pip install "git+https://github.com/AbstractEyes/geolip-core.git"
"""

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as T
import math

try:
    from geolip_core.linalg import svd as geolip_svd
    HAS_GEOLIP = True
    print("Using geolip-core SVD (Gram + eigh)")
except ImportError:
    HAS_GEOLIP = False
    print("geolip-core not found, fallback to torch.svd_lowrank")


# -- CM monitoring --

def cayley_menger_vol2(points):
    B, N, D = points.shape
    gram = torch.bmm(points, points.transpose(1, 2))
    norms = torch.diagonal(gram, dim1=1, dim2=2)
    d2 = F.relu(norms.unsqueeze(2) + norms.unsqueeze(1) - 2 * gram)
    cm = torch.zeros(B, N + 1, N + 1, device=points.device, dtype=points.dtype)
    cm[:, 0, 1:] = 1.0
    cm[:, 1:, 0] = 1.0
    cm[:, 1:, 1:] = d2
    k = N - 1
    sign = (-1.0) ** (k + 1)
    fact = math.factorial(k)
    return sign * torch.linalg.det(cm.float()).to(points.dtype) / ((2 ** k) * (fact ** 2))


def cv_of(emb, n_samples=200):
    if emb.dim() != 2 or emb.shape[0] < 5:
        return 0.0
    N, D = emb.shape
    pool = min(N, 512)
    indices = torch.stack([torch.randperm(pool, device=emb.device)[:5] for _ in range(n_samples)])
    vol2 = cayley_menger_vol2(emb[:pool][indices])
    valid = vol2 > 1e-20
    if valid.sum() < 10:
        return 0.0
    vols = vol2[valid].sqrt()
    return (vols.std() / (vols.mean() + 1e-8)).item()


def safe_gram_svd(M, eps=1e-6):
    """Gram + eigh SVD with diagonal regularization for conditioning."""
    with torch.amp.autocast('cuda', enabled=False):
        A = M.float()
        G = torch.bmm(A.transpose(1, 2), A)
        # Regularize: add eps to diagonal for numerical stability
        G = G + eps * torch.eye(G.shape[-1], device=G.device, dtype=G.dtype).unsqueeze(0)
        eigenvalues, V = torch.linalg.eigh(G)
        eigenvalues = eigenvalues.flip(-1)
        V = V.flip(-1)
        S = torch.sqrt(eigenvalues.clamp(min=1e-12))
        U = torch.bmm(A, V) / S.unsqueeze(1).clamp(min=1e-8)
        Vh = V.transpose(-2, -1).contiguous()
    return U, S, Vh


BINDING_CONSTANT = 0.29154


# -- Data --

def get_cifar10(batch_size=256):
    transform = T.Compose([
        T.ToTensor(),
        T.Normalize((0.4914, 0.4822, 0.4465), (0.2470, 0.2435, 0.2616)),
    ])
    train_ds = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform)
    test_ds = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform)
    train_loader = torch.utils.data.DataLoader(train_ds, batch_size=batch_size, shuffle=True, num_workers=2)
    test_loader = torch.utils.data.DataLoader(test_ds, batch_size=batch_size, shuffle=False, num_workers=2)
    return train_loader, test_loader


# -- SVAE --

class SVAE(nn.Module):
    def __init__(self, matrix_v=1024, D=24):
        super().__init__()
        self.matrix_v = matrix_v
        self.D = D
        self.img_dim = 3 * 32 * 32
        self.mat_dim = matrix_v * D

        # Deep encoder for 1024x24 = 24,576 output
        self.encoder = nn.Sequential(
            nn.Linear(self.img_dim, 1024),
            nn.GELU(),
            nn.Linear(1024, 2048),
            nn.GELU(),
            nn.Linear(2048, self.mat_dim),
        )
        self.decoder = nn.Sequential(
            nn.Linear(self.mat_dim, 2048),
            nn.GELU(),
            nn.Linear(2048, 1024),
            nn.GELU(),
            nn.Linear(1024, self.img_dim),
        )

        # Spectral log-variance (shape regularization only)
        self.logvar_head = nn.Sequential(
            nn.Linear(2048, 128),  # tap from encoder hidden, not full mat_dim
            nn.GELU(),
            nn.Linear(128, D),
        )

        # Prior: SHAPE only, not magnitude
        # Normalized decay from 1.0 to ~0.14 in log space
        # The prior says "S should decay smoothly" not "S should be small"
        self.register_buffer('prior_log_mu', torch.linspace(0, -2, D))
        self.register_buffer('prior_log_var', torch.ones(D))  # wide prior (var=e^1 ~2.7)

    def encode(self, images):
        B = images.shape[0]
        flat = images.reshape(B, -1)

        # Run encoder with hidden tap for logvar
        h1 = F.gelu(self.encoder[0](flat))      # 3072 -> 1024
        h2 = F.gelu(self.encoder[2](h1))         # 1024 -> 2048
        mat_flat = self.encoder[4](h2)            # 2048 -> mat_dim
        M = mat_flat.reshape(B, self.matrix_v, self.D)

        if HAS_GEOLIP:
            try:
                U, S, Vh = geolip_svd(M)
            except Exception:
                U, S, Vh = safe_gram_svd(M)
        else:
            U, S, Vh = safe_gram_svd(M)

        # Log-variance from hidden (not full mat_dim - too expensive)
        log_var = self.logvar_head(h2)

        # Reparameterize on NORMALIZED spectrum (shape, not magnitude)
        if self.training:
            S_norm = S / (S[:, 0:1] + 1e-8)  # normalize by S[0]
            log_S_norm = torch.log(S_norm.clamp(min=1e-8))
            std = torch.exp(0.5 * log_var)
            eps = torch.randn_like(std)
            S_norm_sampled = torch.exp(log_S_norm + std * eps)
            S_norm_sampled, _ = S_norm_sampled.sort(dim=-1, descending=True)
            # Denormalize back
            S_sampled = S_norm_sampled * S[:, 0:1]
        else:
            S_sampled = S
            S_norm = S / (S[:, 0:1] + 1e-8)

        return {
            'U': U, 'S': S, 'S_sampled': S_sampled, 'Vt': Vh,
            'S_norm': S / (S[:, 0:1] + 1e-8),
            'M': M, 'log_var': log_var,
        }

    def decode_from_svd(self, U, S, Vt):
        B = U.shape[0]
        M_hat = torch.bmm(U * S.unsqueeze(1), Vt)
        return self.decoder(M_hat.reshape(B, -1)).reshape(B, 3, 32, 32)

    def spectral_kl(self, S_norm, log_var):
        """KL on NORMALIZED spectrum shape. Magnitude-free."""
        log_S = torch.log(S_norm.clamp(min=1e-8))
        mu_q = log_S
        var_q = torch.exp(log_var)
        mu_p = self.prior_log_mu.unsqueeze(0)
        var_p = torch.exp(self.prior_log_var).unsqueeze(0)
        kl = 0.5 * (var_q / var_p + (mu_p - mu_q).pow(2) / var_p
                     - 1 + torch.log(var_p / (var_q + 1e-8)))
        return kl.sum(dim=-1).mean()

    def forward(self, images):
        svd = self.encode(images)
        recon = self.decode_from_svd(svd['U'], svd['S_sampled'], svd['Vt'])
        kl = self.spectral_kl(svd['S_norm'], svd['log_var'])
        return {'recon': recon, 'svd': svd, 'kl': kl}

    @staticmethod
    def effective_rank(S):
        p = S / (S.sum(-1, keepdim=True) + 1e-8)
        p = p.clamp(min=1e-8)
        return (-(p * p.log()).sum(-1)).exp()


# -- Training --

def train(epochs=50, lr=1e-3, kl_weight=0.001, device='cuda'):
    device = torch.device(device if torch.cuda.is_available() else 'cpu')
    train_loader, test_loader = get_cifar10(batch_size=256)

    model = SVAE(matrix_v=1024, D=24).to(device)
    opt = torch.optim.Adam(model.parameters(), lr=lr)
    sched = torch.optim.lr_scheduler.CosineAnnealingLR(opt, T_max=epochs)

    total_params = sum(p.numel() for p in model.parameters())
    print(f"SVAE - V=1024, D=24 (Validated: CV=0.2916)")
    print(f"  Matrix: (1024, 24) = 24,576 elements")
    print(f"  KL on normalized spectrum shape (weight={kl_weight})")
    print(f"  Wide prior (var=e, shape-only)")
    print(f"  Params: {total_params:,}")
    print("=" * 90)
    print(f"{'ep':>3} | {'loss':>7} {'recon':>7} {'kl':>7} | "
          f"{'t_rec':>7} | "
          f"{'S0':>7} {'SD':>6} {'ratio':>5} {'erank':>5} | "
          f"{'row_cv':>7} {'Sn_shape':>20}")
    print("-" * 90)

    for epoch in range(1, epochs + 1):
        model.train()
        total_loss, total_recon, total_kl, n = 0, 0, 0, 0

        for images, labels in train_loader:
            images = images.to(device)
            opt.zero_grad()
            out = model(images)

            recon_loss = F.mse_loss(out['recon'], images)
            kl = out['kl']
            loss = recon_loss + kl_weight * kl
            loss.backward()
            opt.step()

            total_loss += loss.item() * len(images)
            total_recon += recon_loss.item() * len(images)
            total_kl += kl.item() * len(images)
            n += len(images)

        sched.step()

        if epoch % 2 == 0 or epoch <= 3:
            model.eval()
            test_recon, test_n = 0, 0
            test_S, test_erank = None, 0
            row_cvs = []
            test_S_norm = None
            nb = 0

            with torch.no_grad():
                for images, labels in test_loader:
                    images = images.to(device)
                    out = model(images)
                    test_recon += F.mse_loss(out['recon'], images).item() * len(images)
                    test_n += len(images)
                    test_erank += model.effective_rank(out['svd']['S']).mean().item()

                    if nb < 3:
                        for b in range(min(4, len(images))):
                            row_cvs.append(cv_of(out['svd']['M'][b]))

                    if test_S is None:
                        test_S = out['svd']['S'].mean(0).cpu()
                        test_S_norm = out['svd']['S_norm'].mean(0).cpu()
                    else:
                        test_S += out['svd']['S'].mean(0).cpu()
                        test_S_norm += out['svd']['S_norm'].mean(0).cpu()
                    nb += 1

            test_erank /= nb
            test_S /= nb
            test_S_norm /= nb
            ratio = (test_S[0] / (test_S[-1] + 1e-8)).item()
            mean_cv = sum(row_cvs) / len(row_cvs) if row_cvs else 0

            # Show normalized spectrum shape (first 5)
            shape_str = " ".join(f"{s:.3f}" for s in test_S_norm[:5])

            print(f"{epoch:3d} | {total_loss/n:7.4f} {total_recon/n:7.4f} "
                  f"{total_kl/n:7.3f} | "
                  f"{test_recon/test_n:7.4f} | "
                  f"{test_S[0]:7.2f} {test_S[-1]:6.3f} {ratio:5.2f} "
                  f"{test_erank:5.2f} | "
                  f"{mean_cv:7.4f} [{shape_str}]")

    # -- Final Analysis --
    print()
    print("=" * 90)
    print("FINAL ANALYSIS")
    print("=" * 90)

    model.eval()
    all_S, all_recon_err, all_labels = [], [], []
    all_row_cvs = []

    with torch.no_grad():
        for images, labels in test_loader:
            images = images.to(device)
            out = model(images)
            all_S.append(out['svd']['S'].cpu())
            all_recon_err.append(
                F.mse_loss(out['recon'], images, reduction='none')
                .mean(dim=(1, 2, 3)).cpu())
            all_labels.append(labels.cpu())
            for b in range(min(4, len(images))):
                all_row_cvs.append(cv_of(out['svd']['M'][b]))

    all_S = torch.cat(all_S)
    all_recon_err = torch.cat(all_recon_err)
    all_labels = torch.cat(all_labels)
    erank = model.effective_rank(all_S)
    mean_cv = sum(all_row_cvs) / len(all_row_cvs)

    print(f"\n  V=1024, D=24 (validated CV=0.2916)")
    print(f"  Recon MSE: {all_recon_err.mean():.6f} +/- {all_recon_err.std():.6f}")
    print(f"  Effective rank: {erank.mean():.2f} +/- {erank.std():.2f}")
    print(f"  Row CV: {mean_cv:.4f} (target: {BINDING_CONSTANT}, delta: {abs(mean_cv - BINDING_CONSTANT):.4f})")

    # Spectrum
    S_mean = all_S.mean(0)
    S_norm = S_mean / (S_mean[0] + 1e-8)
    total_energy = (S_mean ** 2).sum()
    print(f"\n  Singular value profile (raw and normalized):")
    cumulative = 0
    for i in range(len(S_mean)):
        e = (S_mean[i] ** 2).item()
        cumulative += e
        pct = cumulative / total_energy * 100
        bar = "#" * int(S_norm[i].item() * 30)
        print(f"    S[{i:2d}]: {S_mean[i]:8.3f}  norm={S_norm[i]:.4f}  cum={pct:5.1f}%  {bar}")

    # Per-class
    cifar_names = ['plane', 'car', 'bird', 'cat', 'deer',
                   'dog', 'frog', 'horse', 'ship', 'truck']
    print(f"\n  Per-class:")
    print(f"    {'cls':>6}  {'recon':>8}  {'erank':>6}  {'S0':>7}  {'SD':>7}  {'ratio':>6}")
    for c in range(10):
        mask = all_labels == c
        rc = all_recon_err[mask].mean().item()
        er = erank[mask].mean().item()
        s0 = all_S[mask, 0].mean().item()
        sd = all_S[mask, -1].mean().item()
        r = s0 / (sd + 1e-8)
        print(f"    {cifar_names[c]:>6}  {rc:8.6f}  {er:6.2f}  {s0:7.3f}  {sd:7.3f}  {r:6.2f}")

    # -- Recon grid --
    print(f"\n  Saving reconstruction grid...")
    import matplotlib
    matplotlib.use('Agg')
    import matplotlib.pyplot as plt

    mean_t = torch.tensor([0.4914, 0.4822, 0.4465]).reshape(1, 3, 1, 1).to(device)
    std_t = torch.tensor([0.2470, 0.2435, 0.2616]).reshape(1, 3, 1, 1).to(device)

    model.eval()
    with torch.no_grad():
        images, labels = next(iter(test_loader))
        images = images.to(device)
        out = model(images)

        selected_idx = []
        for c in range(10):
            class_idx = (labels == c).nonzero(as_tuple=True)[0]
            selected_idx.extend(class_idx[:2].tolist())

        orig = images[selected_idx]
        U = out['svd']['U'][selected_idx]
        S = out['svd']['S'][selected_idx]
        Vt = out['svd']['Vt'][selected_idx]

        mode_counts = [1, 4, 8, 16, 24]
        prog_recons = []
        for nm in mode_counts:
            r = model.decode_from_svd(U[:, :, :nm], S[:, :nm], Vt[:, :nm, :])
            prog_recons.append(r)

    def denorm(t):
        return (t * std_t + mean_t).clamp(0, 1).cpu()

    n_samples = len(selected_idx)
    n_cols = 2 + len(mode_counts)
    fig, axes = plt.subplots(n_samples, n_cols, figsize=(n_cols * 1.5, n_samples * 1.5))
    col_titles = ['Original'] + [f'{m} modes' for m in mode_counts] + ['|Err|x5']

    for i in range(n_samples):
        axes[i, 0].imshow(denorm(orig[i:i+1])[0].permute(1, 2, 0).numpy())
        for j, r in enumerate(prog_recons):
            axes[i, j+1].imshow(denorm(r[i:i+1])[0].permute(1, 2, 0).numpy())
        err_col = 1 + len(prog_recons)
        diff = (denorm(orig[i:i+1]) - denorm(prog_recons[-1][i:i+1])).abs() * 5
        axes[i, err_col].imshow(diff.clamp(0, 1)[0].permute(1, 2, 0).numpy())
        c = labels[selected_idx[i]].item()
        axes[i, 0].set_ylabel(cifar_names[c], fontsize=8, rotation=0, labelpad=35)

    for j, title in enumerate(col_titles):
        axes[0, j].set_title(title, fontsize=8)
    for ax in axes.flat:
        ax.axis('off')

    plt.tight_layout()
    plt.savefig('/content/svae_recon_grid.png', dpi=200, bbox_inches='tight')
    print(f"  Saved to /content/svae_recon_grid.png")
    try:
        plt.show()
    except:
        pass
    plt.close()


if __name__ == "__main__":
    train()