Lipschitz Forward-Invariant Policy Certification / results.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "seed": 2446,
  3  "bound_sweep": [
  4    {
  5      "k": 0.0,
  6      "eps": 0.01,
  7      "predicted_delta": 0.09399999999999999,
  8      "observed_max": 0.09377241407944241,
  9      "ratio_observed_to_bound": 0.9975788731855577
 10    },
 11    {
 12      "k": 0.0,
 13      "eps": 0.08,
 14      "predicted_delta": 0.16399999999999998,
 15      "observed_max": 0.16328784810243863,
 16      "ratio_observed_to_bound": 0.9956576103807234
 17    },
 18    {
 19      "k": 0.0,
 20      "eps": 0.2,
 21      "predicted_delta": 0.28400000000000003,
 22      "observed_max": 0.283197652384377,
 23      "ratio_observed_to_bound": 0.9971748323393556
 24    },
 25    {
 26      "k": 0.2,
 27      "eps": 0.01,
 28      "predicted_delta": 0.1132,
 29      "observed_max": 0.11310145603818444,
 30      "ratio_observed_to_bound": 0.9991294703019827
 31    },
 32    {
 33      "k": 0.2,
 34      "eps": 0.08,
 35      "predicted_delta": 0.1832,
 36      "observed_max": 0.18225589847785584,
 37      "ratio_observed_to_bound": 0.9948466074118768
 38    },
 39    {
 40      "k": 0.2,
 41      "eps": 0.2,
 42      "predicted_delta": 0.3032,
 43      "observed_max": 0.3015575421877501,
 44      "ratio_observed_to_bound": 0.9945829227828169
 45    },
 46    {
 47      "k": 0.5,
 48      "eps": 0.01,
 49      "predicted_delta": 0.14200000000000002,
 50      "observed_max": 0.1419705479307598,
 51      "ratio_observed_to_bound": 0.9997925910616886
 52    },
 53    {
 54      "k": 0.5,
 55      "eps": 0.08,
 56      "predicted_delta": 0.21200000000000002,
 57      "observed_max": 0.21099240248818218,
 58      "ratio_observed_to_bound": 0.995247181548029
 59    },
 60    {
 61      "k": 0.5,
 62      "eps": 0.2,
 63      "predicted_delta": 0.332,
 64      "observed_max": 0.33079045898722004,
 65      "ratio_observed_to_bound": 0.9963568041783736
 66    },
 67    {
 68      "k": 1.0,
 69      "eps": 0.01,
 70      "predicted_delta": 0.19,
 71      "observed_max": 0.1899085031667766,
 72      "ratio_observed_to_bound": 0.9995184377198768
 73    },
 74    {
 75      "k": 1.0,
 76      "eps": 0.08,
 77      "predicted_delta": 0.26,
 78      "observed_max": 0.2591355623578626,
 79      "ratio_observed_to_bound": 0.996675239837933
 80    },
 81    {
 82      "k": 1.0,
 83      "eps": 0.2,
 84      "predicted_delta": 0.38,
 85      "observed_max": 0.37790929177561805,
 86      "ratio_observed_to_bound": 0.9944981362516264
 87    },
 88    {
 89      "k": 1.5,
 90      "eps": 0.01,
 91      "predicted_delta": 0.23800000000000002,
 92      "observed_max": 0.2372067281682338,
 93      "ratio_observed_to_bound": 0.9966669250766125
 94    },
 95    {
 96      "k": 1.5,
 97      "eps": 0.08,
 98      "predicted_delta": 0.308,
 99      "observed_max": 0.30690792028853275,
100      "ratio_observed_to_bound": 0.9964542866510804
101    },
102    {
103      "k": 1.5,
104      "eps": 0.2,
105      "predicted_delta": 0.42800000000000005,
106      "observed_max": 0.42482445365246735,
107      "ratio_observed_to_bound": 0.9925804991880077
108    }
109  ],
110  "boundary_sweep": [
111    {
112      "k": 0.0,
113      "predicted_eps_boundary": 0.916,
114      "observed_eps_boundary": 0.92,
115      "relative_error": 0.00436681222707424
116    },
117    {
118      "k": 0.25,
119      "predicted_eps_boundary": 0.892,
120      "observed_eps_boundary": 0.895,
121      "relative_error": 0.0033632286995515723
122    },
123    {
124      "k": 0.6,
125      "predicted_eps_boundary": 0.8584,
126      "observed_eps_boundary": 0.86,
127      "relative_error": 0.001863932898415581
128    },
129    {
130      "k": 1.0,
131      "predicted_eps_boundary": 0.8200000000000001,
132      "observed_eps_boundary": 0.8200000000000001,
133      "relative_error": 0.0
134    }
135  ],
136  "gain_scaling": [
137    {
138      "k": 0.0,
139      "margin": 0.866,
140      "margin_drop": 0.0,
141      "predicted_linear_drop": 0.0
142    },
143    {
144      "k": 0.25,
145      "margin": 0.8420000000000001,
146      "margin_drop": 0.02399999999999991,
147      "predicted_linear_drop": 0.024
148    },
149    {
150      "k": 0.5,
151      "margin": 0.8180000000000001,
152      "margin_drop": 0.04799999999999993,
153      "predicted_linear_drop": 0.048
154    },
155    {
156      "k": 0.75,
157      "margin": 0.794,
158      "margin_drop": 0.07199999999999995,
159      "predicted_linear_drop": 0.07200000000000001
160    },
161    {
162      "k": 1.0,
163      "margin": 0.77,
164      "margin_drop": 0.09599999999999997,
165      "predicted_linear_drop": 0.096
166    }
167  ],
168  "training": {
169    "baseline": {
170      "lipschitz_upper_bound": 1.536645583956897,
171      "certified_margin": 0.7184820239401379,
172      "nominal_violation_mean": 0.011351236142218113,
173      "empirical_action_sensitivity": 0.5853400230407715
174    },
175    "certified": {
176      "lipschitz_upper_bound": 1.536645583956897,
177      "certified_margin": 0.7184820239401379,
178      "nominal_violation_mean": 0.011351236142218113,
179      "empirical_action_sensitivity": 0.5853400230407715
180    }
181  }
182}