ISS-Gated Positive Neural State Module / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.005020207085181028
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.002,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.004117600648896769
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0029489120497601107
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.001,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.005006968247471377
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.002,
 44          "weight_decay": 0.0001
 45        },
 46        "mean": 0.004086799017386511
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.0029277410794747993
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0029277410794747993,
 58      "std": 0.0017676469099389427,
 59      "per_seed": [
 60        0.00244312291033566,
 61        0.002742342883720994,
 62        0.0011618298012763262,
 63        0.007381403353065252,
 64        0.00252148206345737,
 65        0.003086549462750554,
 66        0.002133725443854928,
 67        0.0019514727173373103
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.0024121131718857214,
 74    "std": 0.0011804587505530163,
 75    "per_seed": [
 76      0.002581692300736904,
 77      0.002109199995175004,
 78      0.001057643094100058,
 79      0.005024303216487169,
 80      0.0026745826471596956,
 81      0.0028591060545295477,
 82      0.0019529175478965044,
 83      0.0010374605190008879
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": -0.000515627907589078,
 89    "idea_wins": 6,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      0.00013856939040124416,
 93      -0.00063314288854599,
 94      -0.0001041867071762681,
 95      -0.002357100136578083,
 96      0.00015310058370232582,
 97      -0.0002274434082210064,
 98      -0.00018080789595842361,
 99      -0.0009140121983364224
100    ],
101    "p_value": 0.0757,
102    "mde": 0.0006926282795900552,
103    "mde_rel_pct": 23.65742942385821,
104    "verdict": "no significant win",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "prediction": {
109      "claim": "bounded positive state should remain finite and ISS energy should not grow",
110      "baseline_pred_abs_mean": 0.714909203350544,
111      "observed_Vdot_mean": 1.7621822357177734,
112      "observed_V_mean": 0.023558731889352202,
113      "observed_state_min": 0.05470801889896393,
114      "observed_state_max": 3.6606810092926025,
115      "interpretation": "Positive finite states were observed, but mean Vdot was positive rather than nonpositive; the ISS decay prediction is not confirmed."
116    },
117    "confirmed": false
118  },
119  "idea_sweep": [
120    {
121      "cfg": {
122        "lr": 0.001,
123        "weight_decay": 0.0,
124        "lambda_iss": 0.0
125      },
126      "mean": 0.14595716912299395
127    },
128    {
129      "cfg": {
130        "lr": 0.002,
131        "weight_decay": 0.0,
132        "lambda_iss": 0.01
133      },
134      "mean": 0.0060275343275861815
135    },
136    {
137      "cfg": {
138        "lr": 0.003,
139        "weight_decay": 0.0001,
140        "lambda_iss": 0.03
141      },
142      "mean": 0.0024121131718857214
143    }
144  ]
145}