Add comprehensive evaluation and ablation runner

example/window_models.py

@@ -0,0 +1,65 @@
+#!/usr/bin/env python3
+"""Small neural models used by the evaluation suite."""
+
+from __future__ import annotations
+
+import torch
+import torch.nn as nn
+
+
+class MLPClassifier(nn.Module):
+    def __init__(self, input_dim: int, hidden_dim: int = 256, dropout: float = 0.1):
+        super().__init__()
+        mid_dim = max(hidden_dim // 2, 32)
+        self.net = nn.Sequential(
+            nn.Linear(input_dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, mid_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(mid_dim, 1),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.net(x).squeeze(-1)
+
+
+class MLPRegressor(nn.Module):
+    def __init__(self, input_dim: int, output_dim: int, hidden_dim: int = 256, dropout: float = 0.1):
+        super().__init__()
+        mid_dim = max(hidden_dim // 2, 32)
+        self.net = nn.Sequential(
+            nn.Linear(input_dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, mid_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(mid_dim, output_dim),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.net(x)
+
+
+class MLPAutoencoder(nn.Module):
+    def __init__(self, input_dim: int, hidden_dim: int = 256, latent_dim: int = 64, dropout: float = 0.1):
+        super().__init__()
+        self.encoder = nn.Sequential(
+            nn.Linear(input_dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, latent_dim),
+            nn.GELU(),
+        )
+        self.decoder = nn.Sequential(
+            nn.Linear(latent_dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, input_dim),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.decoder(self.encoder(x))
+