results: validate Rain layer-state mechanism

1 files changed, 130 insertions, 0 deletions

diff --git a/results/ep_bias/s1_summary.json b/results/ep_bias/s1_summary.json
new file mode 100644
index 0000000..dc43021
--- /dev/null
+++ b/results/ep_bias/s1_summary.json
@@ -0,0 +1,130 @@
+{
+  "matched_predictor_protocol": {
+    "constant_neutral_observations": 128,
+    "extra_equilibrium_phases": 0,
+    "innovation_neutral_observations": 128,
+    "predictor_updates_after_first_minibatch": 0,
+    "source": "existing first EP phase of the first training minibatch"
+  },
+  "paired_development_effects": {
+    "clean_minus_innovation_accuracy_points": 0.9499999999999953,
+    "innovation_minus_constant_accuracy_points": 14.150000000000007,
+    "innovation_minus_raw_accuracy_points": 27.800000000000004,
+    "innovation_wall_over_raw_ratio": 1.0050329418868948,
+    "raw_loss_recovered_fraction": 0.9669565217391306
+  },
+  "ratio": 0.01,
+  "rows": {
+    "clean": {
+      "all_finite": true,
+      "final_corrector": {},
+      "final_test_accuracy": 0.7805,
+      "neutral_observations": 0,
+      "wall_seconds": 32.770522356033325
+    },
+    "constant": {
+      "all_finite": true,
+      "final_corrector": {
+        "bias_state_difference_rms": 7.489153969713887e-05,
+        "bias_to_clean_state_difference_rms": 0.03071981568073833,
+        "clean_state_difference_rms": 0.002437890268465924,
+        "neutral_observations": 128,
+        "residual_state_difference_rms": 7.972976077708221e-05,
+        "residual_to_clean_state_difference_rms": 0.03270440913948652,
+        "step": 236
+      },
+      "final_test_accuracy": 0.6295,
+      "neutral_observations": 128,
+      "wall_seconds": 32.627506256103516
+    },
+    "innovation": {
+      "all_finite": true,
+      "final_corrector": {
+        "bias_state_difference_rms": 7.2471939856954e-05,
+        "bias_to_clean_state_difference_rms": 0.027095526020707417,
+        "clean_state_difference_rms": 0.002674682890509977,
+        "neutral_observations": 128,
+        "residual_state_difference_rms": 3.644143814163778e-05,
+        "residual_to_clean_state_difference_rms": 0.013624582663961916,
+        "step": 236
+      },
+      "final_test_accuracy": 0.771,
+      "neutral_observations": 128,
+      "wall_seconds": 32.43295383453369
+    },
+    "oracle": {
+      "all_finite": true,
+      "final_corrector": {
+        "bias_state_difference_rms": 7.640068624454531e-05,
+        "bias_to_clean_state_difference_rms": 0.024389158553879114,
+        "clean_state_difference_rms": 0.0031325675330604514,
+        "neutral_observations": 0,
+        "residual_state_difference_rms": 0.0,
+        "residual_to_clean_state_difference_rms": 0.0,
+        "step": 236
+      },
+      "final_test_accuracy": 0.781,
+      "neutral_observations": 0,
+      "wall_seconds": 33.032013177871704
+    },
+    "raw": {
+      "all_finite": true,
+      "final_corrector": {
+        "bias_state_difference_rms": 6.673162402742633e-05,
+        "bias_to_clean_state_difference_rms": 0.026694932470610805,
+        "clean_state_difference_rms": 0.0024997862085207757,
+        "neutral_observations": 0,
+        "residual_state_difference_rms": 6.673162402742633e-05,
+        "residual_to_clean_state_difference_rms": 0.026694932470610805,
+        "step": 236
+      },
+      "final_test_accuracy": 0.493,
+      "neutral_observations": 0,
+      "wall_seconds": 32.270538091659546
+    },
+    "same_rms_noise": {
+      "all_finite": true,
+      "final_corrector": {
+        "bias_state_difference_rms": 7.650330356103333e-05,
+        "bias_to_clean_state_difference_rms": 0.026813529453929254,
+        "clean_state_difference_rms": 0.0028531605170621257,
+        "neutral_observations": 0,
+        "residual_state_difference_rms": 7.797371328954984e-05,
+        "residual_to_clean_state_difference_rms": 0.027328891179890114,
+        "step": 236
+      },
+      "final_test_accuracy": 0.7765,
+      "neutral_observations": 0,
+      "wall_seconds": 31.844523668289185
+    }
+  },
+  "stage": "rain_ep_layer_state_s1",
+  "status": "positive_single_seed_development_not_confirmation",
+  "stronger_ratio_boundary": {
+    "r0p1": {
+      "constant": {
+        "all_finite": true,
+        "epochs_completed": 3,
+        "final_test_accuracy": 0.0985
+      },
+      "innovation": {
+        "all_finite": true,
+        "epochs_completed": 3,
+        "final_test_accuracy": 0.0975
+      }
+    },
+    "r4": {
+      "constant": {
+        "all_finite": false,
+        "epochs_completed": 1,
+        "final_test_accuracy": 0.1
+      },
+      "innovation": {
+        "all_finite": false,
+        "epochs_completed": 1,
+        "final_test_accuracy": 0.1
+      }
+    }
+  },
+  "test_policy": "development test subset observed each epoch; ratio chosen here; all accuracy claims require a new frozen confirmation"
+}