experiments: freeze oral A development gates

1 files changed, 65 insertions, 0 deletions

diff --git a/experiments/oral_a_short_screen.py b/experiments/oral_a_short_screen.py
new file mode 100755
index 0000000..51df778
--- /dev/null
+++ b/experiments/oral_a_short_screen.py
@@ -0,0 +1,65 @@
+#!/usr/bin/env python3
+"""Run a deterministic shard of the frozen A2b short accuracy grid."""
+import argparse
+import json
+import os
+import subprocess
+import sys
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--selection", default="results/oral_a_apical_selection.json")
+    parser.add_argument("--device", default="cuda")
+    parser.add_argument("--shard_index", type=int, default=0)
+    parser.add_argument("--num_shards", type=int, default=1)
+    parser.add_argument("--dry_run", action="store_true")
+    args = parser.parse_args()
+    if not 0 <= args.shard_index < args.num_shards:
+        raise ValueError("invalid shard index")
+    with open(args.selection) as handle:
+        selection = json.load(handle)
+    if selection["status"] != "selected":
+        raise ValueError("A2a did not select both vectorizer families")
+
+    common = [
+        sys.executable, "experiments/conv_run.py", "--device", args.device,
+        "--depth", "20", "--width", "16", "--seed", "0", "--loader_seed", "0",
+        "--batch_size", "128", "--epochs", "20", "--train_limit", "10000",
+        "--val_examples", "5000", "--split_seed", "2027",
+        "--eval_split", "validation", "--eval_every", "0", "--augment_train", "1",
+        "--lr_schedule", "cosine", "--warmup_epochs", "0", "--momentum", "0.9",
+        "--weight_decay", "1e-4", "--normalization", "batchnorm",
+    ]
+    jobs = []
+    jobs.append(("bp_lr0.1", common + [
+        "--mode", "bp", "--lr", "0.1",
+        "--out", "results/oral_a_short/bp_lr0.1.json"]))
+    for rate in (0.01, 0.03, 0.1):
+        jobs.append((f"dfa_lr{rate}", common + [
+            "--mode", "dfa", "--lr", str(rate), "--output_lr", "0.1",
+            "--a_scale", "1", "--vectorizer_mode", "spatial_template",
+            "--out", f"results/oral_a_short/dfa_lr{rate}.json"]))
+    for mode in ("spatial_template", "channel_gated"):
+        chosen = selection["selected"][mode]
+        for rate in (0.01, 0.03, 0.1):
+            tag = f"sdil_{mode}_lr{rate}"
+            jobs.append((tag, common + [
+                "--mode", "sdil", "--lr", str(rate), "--output_lr", "0.1",
+                "--vectorizer_mode", mode,
+                "--a_scale", str(chosen["a_scale"]),
+                "--eta_A", str(chosen["eta_A"]), "--a_warmup_steps", "100",
+                "--pert_sigma", "0.01", "--pert_directions", "1",
+                "--pert_every", "4", "--alignment_probe", "32",
+                "--out", f"results/oral_a_short/{tag}.json"]))
+    os.makedirs("results/oral_a_short", exist_ok=True)
+    selected_jobs = [job for index, job in enumerate(jobs)
+                     if index % args.num_shards == args.shard_index]
+    for tag, command in selected_jobs:
+        print(tag, " ".join(command), flush=True)
+        if not args.dry_run:
+            subprocess.run(command, check=True)
+
+
+if __name__ == "__main__":
+    main()