Noise-Whitened Trajectory-KL Policy Regularization / results.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "seed": 1267,
  3  "math_check": {
  4    "stationarity_gradient_norm": 6.938893903907227e-13,
  5    "hessian_error_norm": 1.5476311098377395e-07,
  6    "analytic_u": [
  7      0.14427515036590355,
  8      -0.24895276547974993
  9    ]
 10  },
 11  "toy_predictions": {
 12    "lambdas": [
 13      0.0,
 14      0.1,
 15      0.3,
 16      1.0,
 17      3.0,
 18      10.0,
 19      30.0,
 20      100.0
 21    ],
 22    "solutions": [
 23      [
 24        2.0,
 25        2.0
 26      ],
 27      [
 28        1.4285714285714286,
 29        1.9512195121951221
 30      ],
 31      [
 32        0.9090909090909091,
 33        1.8604651162790697
 34      ],
 35      [
 36        0.4,
 37        1.6
 38      ],
 39      [
 40        0.15384615384615385,
 41        1.1428571428571428
 42      ],
 43      [
 44        0.04878048780487805,
 45        0.5714285714285714
 46      ],
 47      [
 48        0.01652892561983471,
 49        0.23529411764705882
 50      ],
 51      [
 52        0.004987531172069825,
 53        0.07692307692307693
 54      ]
 55    ],
 56    "trajectory_kl": [
 57      8.5,
 58      4.557539851157596,
 59      2.0855588680948824,
 60      0.6400000000000001,
 61      0.21060258422895783,
 62      0.04557539851157595,
 63      0.0074668259892055515,
 64      0.0007893959047989377
 65    ],
 66    "prediction_1_kl_monotone": true,
 67    "prediction_2_expected_lambda_u_tail": [
 68      0.5,
 69      8.0
 70    ],
 71    "prediction_2_observed_lambda_u_tail": [
 72      0.4941419212836014,
 73      6.821805645335057
 74    ],
 75    "prediction_2_relative_error": 0.1472742943330995,
 76    "prediction_3_predicted_curvature_ratio": 4.0,
 77    "prediction_3_observed_curvature_ratio": 4.0,
 78    "prediction_3_relative_error": 0.0
 79  },
 80  "policy_experiment": {
 81    "device": "cuda",
 82    "lambda": 2.0,
 83    "action_regularization": {
 84      "task_mse_half": 0.3949859142303467,
 85      "action_penalty": 0.049373235553503036,
 86      "trajectory_kl": 0.1613738238811493,
 87      "W": [
 88        [
 89          0.20000000298023224,
 90          -0.07999999821186066
 91        ],
 92        [
 93          0.03999998793005943,
 94          0.1599999964237213
 95        ],
 96        [
 97          -0.13999998569488525,
 98          0.06000000238418579
 99        ]
100      ]
101    },
102    "noise_whitened_drift_regularization": {
103      "task_mse_half": 0.411968857049942,
104      "action_penalty": 0.11426855623722076,
105      "trajectory_kl": 0.037015583366155624,
106      "W": [
107        [
108          0.0476190522313118,
109          -0.20000000298023224
110        ],
111        [
112          0.00952380895614624,
113          0.4000000059604645
114        ],
115        [
116          -0.03333333507180214,
117          0.15000000596046448
118        ]
119      ]
120    }
121  }
122}