Graded Levy-area recurrent state / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "math": {
  3    "reversal_sweep": [
  4      {
  5        "a": 0.25,
  6        "predicted_abs": 0.125,
  7        "observed_abs": 0.125,
  8        "reverse_sum": 0.0
  9      },
 10      {
 11        "a": 0.5,
 12        "predicted_abs": 0.25,
 13        "observed_abs": 0.25,
 14        "reverse_sum": 0.0
 15      },
 16      {
 17        "a": 1.0,
 18        "predicted_abs": 0.5,
 19        "observed_abs": 0.5,
 20        "reverse_sum": 0.0
 21      },
 22      {
 23        "a": 2.0,
 24        "predicted_abs": 1.0,
 25        "observed_abs": 1.0,
 26        "reverse_sum": 0.0
 27      },
 28      {
 29        "a": 4.0,
 30        "predicted_abs": 2.0,
 31        "observed_abs": 2.0,
 32        "reverse_sum": 0.0
 33      }
 34    ],
 35    "linear_accumulation": [
 36      {
 37        "n": 1,
 38        "predicted": 0.5,
 39        "observed": 0.5
 40      },
 41      {
 42        "n": 2,
 43        "predicted": 1.0,
 44        "observed": 1.0
 45      },
 46      {
 47        "n": 3,
 48        "predicted": 1.5,
 49        "observed": 1.5
 50      },
 51      {
 52        "n": 4,
 53        "predicted": 2.0,
 54        "observed": 2.0
 55      },
 56      {
 57        "n": 5,
 58        "predicted": 2.5,
 59        "observed": 2.5
 60      },
 61      {
 62        "n": 6,
 63        "predicted": 3.0,
 64        "observed": 3.0
 65      },
 66      {
 67        "n": 7,
 68        "predicted": 3.5,
 69        "observed": 3.5
 70      },
 71      {
 72        "n": 8,
 73        "predicted": 4.0,
 74        "observed": 4.0
 75      }
 76    ],
 77    "dilation_sweep": [
 78      {
 79        "lambda": 0.25,
 80        "u_ratio_to_lambda": 1.0,
 81        "A_ratio_to_lambda2": 1.0,
 82        "Q_ratio_to_lambda2": 1.0
 83      },
 84      {
 85        "lambda": 0.5,
 86        "u_ratio_to_lambda": 1.0,
 87        "A_ratio_to_lambda2": 1.0,
 88        "Q_ratio_to_lambda2": 1.0
 89      },
 90      {
 91        "lambda": 1.0,
 92        "u_ratio_to_lambda": 1.0,
 93        "A_ratio_to_lambda2": 1.0,
 94        "Q_ratio_to_lambda2": 1.0
 95      },
 96      {
 97        "lambda": 2.0,
 98        "u_ratio_to_lambda": 1.0,
 99        "A_ratio_to_lambda2": 1.0,
100        "Q_ratio_to_lambda2": 1.0
101      },
102      {
103        "lambda": 4.0,
104        "u_ratio_to_lambda": 1.0,
105        "A_ratio_to_lambda2": 1.0,
106        "Q_ratio_to_lambda2": 1.0
107      }
108    ],
109    "chunk_max_abs_error": 2.6645352591003757e-15
110  },
111  "benchmark": {
112    "additive": {
113      "test_accuracy": 0.51171875,
114      "final_train_loss": 0.6893957257270813,
115      "seconds": 0.2683849334716797,
116      "parameters": 82
117    },
118    "levy": {
119      "test_accuracy": 0.51171875,
120      "final_train_loss": 0.6624104976654053,
121      "seconds": 2.7218849658966064,
122      "parameters": 154
123    },
124    "gru": {
125      "test_accuracy": 0.484375,
126      "final_train_loss": 0.43832412362098694,
127      "seconds": 1.456101894378662,
128      "parameters": 514
129    },
130    "device": "cpu"
131  }
132}