summaryrefslogtreecommitdiff
path: root/results/ep_bias/s3_summary.json
diff options
context:
space:
mode:
authorYurenHao0426 <Blackhao0426@gmail.com>2026-08-06 17:23:24 -0500
committerYurenHao0426 <Blackhao0426@gmail.com>2026-08-06 17:23:24 -0500
commitce1444f31ceab281e59f6aa696131fbcb1c56fc7 (patch)
tree3fdee70019fb3e1e424551034bcd21a1da846ab9 /results/ep_bias/s3_summary.json
parentf631d9709ff695a0015d6cae8b90b3c3802ee9a3 (diff)
results: validate beta-independent Rain bias
Diffstat (limited to 'results/ep_bias/s3_summary.json')
-rw-r--r--results/ep_bias/s3_summary.json144
1 files changed, 144 insertions, 0 deletions
diff --git a/results/ep_bias/s3_summary.json b/results/ep_bias/s3_summary.json
new file mode 100644
index 0000000..47ee830
--- /dev/null
+++ b/results/ep_bias/s3_summary.json
@@ -0,0 +1,144 @@
+{
+ "bias_model": {
+ "depends_on_beta_sign": false,
+ "depends_on_clean_task_difference": false,
+ "normalization": "initial_free_layer_state_rms",
+ "ratio": 0.0002
+ },
+ "effects": {
+ "clean_minus_innovation_accuracy_points": 1.4000000000000012,
+ "constant_final_residual_to_clean_state_difference_rms": 0.02003074134273804,
+ "innovation_final_residual_to_clean_state_difference_rms": 0.0024074442976448505,
+ "innovation_minus_constant_accuracy_points": 10.649999999999993,
+ "innovation_minus_raw_accuracy_points": 15.000000000000002,
+ "innovation_wall_over_clean": 0.9934842384103828
+ },
+ "random_beta_boundary": {
+ "fixed_negative_fails_with_zero_bias": true,
+ "interpretation": "Negative-beta relaxation is not a valid drop-in baseline under the pinned author's positive-EP hyperparameters. Random-sign beta requires separate stability work before comparison; this screen is not an SDIL win over a functioning random-beta method.",
+ "random_sign_fails_with_zero_bias": true
+ },
+ "rows": {
+ "clean": {
+ "beta_sign_counts": {
+ "negative": 0,
+ "positive": 0
+ },
+ "final_corrector": {},
+ "final_validation_accuracy": 0.7915,
+ "finite": true,
+ "wall_seconds": 32.58404016494751
+ },
+ "constant": {
+ "beta_sign_counts": {
+ "negative": 0,
+ "positive": 0
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 6.859923823062401e-05,
+ "bias_to_clean_state_difference_rms": 0.03287761708718417,
+ "clean_state_difference_rms": 0.002086502742845201,
+ "neutral_observations": 128,
+ "residual_state_difference_rms": 4.179419675284569e-05,
+ "residual_to_clean_state_difference_rms": 0.02003074134273804,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.671,
+ "finite": true,
+ "wall_seconds": 32.619558334350586
+ },
+ "fixed_negative_zero_bias": {
+ "beta_sign_counts": {
+ "negative": 237,
+ "positive": 0
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 0.0,
+ "bias_to_clean_state_difference_rms": 0.0,
+ "clean_state_difference_rms": 0.004298488890043378,
+ "neutral_observations": 0,
+ "residual_state_difference_rms": 0.0,
+ "residual_to_clean_state_difference_rms": 0.0,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.099,
+ "finite": true,
+ "wall_seconds": 32.60152792930603
+ },
+ "innovation": {
+ "beta_sign_counts": {
+ "negative": 0,
+ "positive": 0
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 6.27170698976096e-05,
+ "bias_to_clean_state_difference_rms": 0.021524865755300427,
+ "clean_state_difference_rms": 0.0029137031845211728,
+ "neutral_observations": 128,
+ "residual_state_difference_rms": 7.014578116605139e-06,
+ "residual_to_clean_state_difference_rms": 0.0024074442976448505,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.7775,
+ "finite": true,
+ "wall_seconds": 32.3717303276062
+ },
+ "random_beta": {
+ "beta_sign_counts": {
+ "negative": 122,
+ "positive": 115
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 6.120117185015426e-05,
+ "bias_to_clean_state_difference_rms": 0.014596019891031983,
+ "clean_state_difference_rms": 0.004193004141338366,
+ "neutral_observations": 0,
+ "residual_state_difference_rms": 6.120117105019515e-05,
+ "residual_to_clean_state_difference_rms": 0.014596019700247742,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.099,
+ "finite": true,
+ "wall_seconds": 31.93136191368103
+ },
+ "random_beta_zero_bias": {
+ "beta_sign_counts": {
+ "negative": 122,
+ "positive": 115
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 0.0,
+ "bias_to_clean_state_difference_rms": 0.0,
+ "clean_state_difference_rms": 0.00419313641524673,
+ "neutral_observations": 0,
+ "residual_state_difference_rms": 0.0,
+ "residual_to_clean_state_difference_rms": 0.0,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.099,
+ "finite": true,
+ "wall_seconds": 32.36788892745972
+ },
+ "raw": {
+ "beta_sign_counts": {
+ "negative": 0,
+ "positive": 0
+ },
+ "final_corrector": {
+ "bias_state_difference_rms": 7.06939035519467e-05,
+ "bias_to_clean_state_difference_rms": 0.03239290974246007,
+ "clean_state_difference_rms": 0.002182388186612404,
+ "neutral_observations": 0,
+ "residual_state_difference_rms": 7.069390339161565e-05,
+ "residual_to_clean_state_difference_rms": 0.0323929096689942,
+ "step": 236
+ },
+ "final_validation_accuracy": 0.6275,
+ "finite": true,
+ "wall_seconds": 32.71566128730774
+ }
+ },
+ "stage": "rain_ep_beta_independent_hardware_bias_s3",
+ "status": "positive_single_seed_development_negative_beta_unavailable",
+ "test_policy": "new development train holdout data seed 6200"
+}