From 2fa24acae8bb7f8c026db2f7fdade4a29b640d8d Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Wed, 8 Apr 2026 19:24:06 -0500 Subject: Sync experiment+protocol scripts with v2.32 corrected control values MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The pre-v2.31 unsourced values BP=0.609 and DFA=0.308 (which v2.31 fixed to 0.585 and 0.301 via matched 30-ep controls) were also hardcoded as "compare to" comments in 5 helper scripts: experiments/bp_with_penalty_control.py experiments/dfa_residual_penalty_test.py experiments/resmlp_frozen_blocks_baseline.py protocol/examples/threshold_d_sensitivity.py protocol/examples/plot_penalty_rescue.py These are non-paper-input scripts (their output goes to stdout, not to the paper), so the stale values didn't cause numerical errors in the paper itself. But the original v2.31 BP+pen=0.609 unsourced number bug came from exactly this kind of hardcoded "for-comparison" comment that was never measured. Updating them now to remove the same trap from future runs. Each script now references the matched 30-ep 3-seed values from results/bp_no_penalty_30ep, results/dfa_no_penalty_30ep, results/ dfa_pen_short, and results/bp_with_penalty. protocol/EVIDENCE_SUMMARY.md and PAPER_OUTLINE.md still have stale numbers — these are project scratch documents and not user-facing. Deferred to a separate sweep if needed. Co-Authored-By: Claude Opus 4.6 (1M context) --- protocol/examples/plot_penalty_rescue.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'protocol/examples/plot_penalty_rescue.py') diff --git a/protocol/examples/plot_penalty_rescue.py b/protocol/examples/plot_penalty_rescue.py index 37b0fa9..fff300e 100644 --- a/protocol/examples/plot_penalty_rescue.py +++ b/protocol/examples/plot_penalty_rescue.py @@ -10,7 +10,7 @@ Data sources: - vanilla DFA trajectory: results/snapshot_evolution_v2/snapshot_evolution_s42.json - penalized DFA (lam=1e-2): results/dfa_residual_penalty/dfa_pen_lam0.01_s42.json - DFA-shallow baseline 3-seed mean (drawn as horizontal line): 0.349 - - BP-trainable 3-seed mean: 0.609 + - BP-trainable 3-seed mean: 0.6147 (100 ep) / 0.585 (matched 30 ep) Run: python -m protocol.examples.plot_penalty_rescue @@ -91,7 +91,7 @@ def main(): ax.plot([e["epoch"] for e in penalty], [e["acc_eval"] for e in penalty], label=r"DFA + $\lambda \|f_l\|^2$", color="C2", lw=2, marker="o", markersize=4) ax.axhline(0.349, color="k", linestyle="--", lw=1.2, label="DFA-shallow 0.349") - ax.axhline(0.609, color="C0", linestyle=":", lw=1, label="BP-trainable 0.609") + ax.axhline(0.6147, color="C0", linestyle=":", lw=1, label="BP-trainable 100ep 0.615") ax.set_xlabel("epoch", fontsize=10) ax.set_ylabel("test acc", fontsize=10) ax.set_title("(d) headline accuracy", fontsize=11) -- cgit v1.2.3