Add memory centering, grid search experiments, and energy visualizationsHEADmaster

- Add centering support to MemoryBank (center_query, apply_centering, mean
  persistence in save/load) to remove centroid attractor in Hopfield dynamics
- Add center flag to MemoryBankConfig, device field to PipelineConfig
- Grid search scripts: initial (β≤8), residual, high-β, and centered grids
  with dedup-based LLM caching (89-91% call savings)
- Energy landscape visualization: 2D contour, 1D profile, UMAP, PCA heatmap
  comparing centered vs uncentered dynamics
- Experiment log (note.md) documenting 4 rounds of results and root cause
  analysis of centroid attractor problem
- Key finding: β_critical ≈ 37.6 for centered memory; best configs beat
  FAISS baseline by +3-4% F1

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

1 files changed, 2 insertions, 2 deletions

diff --git a/scripts/build_memory_bank.py b/scripts/build_memory_bank.py
index 2aff828..0fc1c51 100644
--- a/scripts/build_memory_bank.py
+++ b/scripts/build_memory_bank.py
@@ -50,13 +50,13 @@ def main() -> None:
     logger.info("Loaded %d passages", len(passages))
 
     # Encode passages in batches
-    encoder = Encoder(encoder_config)
+    encoder = Encoder(encoder_config, device=args.device)
     all_embeddings = []
 
     for i in tqdm(range(0, len(passages), encoder_config.batch_size), desc="Encoding"):
         batch = passages[i : i + encoder_config.batch_size]
         emb = encoder.encode(batch)  # (batch_size, d)
-        all_embeddings.append(emb.cpu())
+        all_embeddings.append(emb.cpu())  # Always store on CPU for saving
 
     embeddings = torch.cat(all_embeddings, dim=0)  # (N, d)
     logger.info("Encoded %d passages -> embeddings shape: %s", len(passages), embeddings.shape)