Incidence-Matrix Structured Action Head / results.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
 1{
 2  "device": "cuda",
 3  "exact_incidence": {
 4    "max_abs_error": 9.5367431640625e-07,
 5    "all_nonempty_rows": true
 6  },
 7  "prediction_1_exact_additivity": "l_p equals sum of incident s_m; numerical max error reported above",
 8  "prediction_2_normalization_scaling": [
 9    {
10      "alpha": 0.0,
11      "predicted_loglog_slope": 1.0,
12      "observed_loglog_slope": 0.9999999999999999,
13      "max_relative_error": 0.0,
14      "logit_k1": 2.0,
15      "logit_k32": 64.0
16    },
17    {
18      "alpha": 0.5,
19      "predicted_loglog_slope": 0.5,
20      "observed_loglog_slope": 0.5000000000000959,
21      "max_relative_error": 5.000444502911705e-13,
22      "logit_k1": 1.999999999999,
23      "logit_k32": 11.313708498984584
24    },
25    {
26      "alpha": 1.0,
27      "predicted_loglog_slope": 0.0,
28      "observed_loglog_slope": 1.9192680039978662e-13,
29      "max_relative_error": 1.000088900582341e-12,
30      "logit_k1": 1.9999999999979998,
31      "logit_k32": 1.9999999999999374
32    }
33  ],
34  "prediction_3_equal-score_size_bias": [
35    {
36      "alpha": 0.0,
37      "odds_k16_vs_k1_observed": 42.52108200006278,
38      "odds_k16_vs_k1_predicted": 42.52108200006278,
39      "odds_all_sizes": [
40        1.0,
41        1.2840254166877414,
42        2.117000016612675,
43        5.754602676005731,
44        42.52108200006278
45      ]
46    },
47    {
48      "alpha": 0.5,
49      "odds_k16_vs_k1_observed": 2.117000016612675,
50      "odds_k16_vs_k1_predicted": 2.117000016612675,
51      "odds_all_sizes": [
52        1.0,
53        1.1091050075586697,
54        1.2840254166877414,
55        1.5794975358658994,
56        2.117000016612675
57      ]
58    },
59    {
60      "alpha": 1.0,
61      "odds_k16_vs_k1_observed": 1.0,
62      "odds_k16_vs_k1_predicted": 1.0,
63      "odds_all_sizes": [
64        1.0,
65        1.0,
66        1.0,
67        1.0,
68        1.0
69      ]
70    }
71  ],
72  "mini_experiment": {
73    "structured": {
74      "unseen_schema_accuracy": 0.828,
75      "unseen_schema_ce": 0.3566694409334799,
76      "parameters": 97,
77      "seconds": 51.7643404006958
78    },
79    "padded_mlp": {
80      "unseen_schema_accuracy": 0.068,
81      "unseen_schema_ce": 13.980102838516235,
82      "parameters": 3136,
83      "seconds": 61.86663246154785
84    },
85    "padding_elements_per_instance": 2048,
86    "incidence_elements_per_instance_example": 168
87  }
88}