File size: 7,461 Bytes
7b513a5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2cafc70
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7b513a5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
{
  "citation": [
    {
      "text": "Trabucco et al. (2022), Design-Bench.",
      "url": "https://arxiv.org/abs/2202.08450"
    },
    {
      "text": "Todorov, Erez, and Tassa (2012), MuJoCo.",
      "url": "https://doi.org/10.1109/IROS.2012.6386109"
    }
  ],
  "context": [
    {
      "constraints": [
        {
          "kind": "range",
          "maximum": 3199,
          "minimum": 0
        }
      ],
      "description": "Maintainer-only row index in the Design-Bench array.",
      "name": "source_index",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 31,
          "minimum": 31
        }
      ],
      "description": "SHA-256-derived identity of canonical float32 weights.",
      "name": "policy_identity",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "finite"
        }
      ],
      "description": "Historical Design-Bench return retained for provenance.",
      "name": "source_return",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 500,
          "minimum": 500
        },
        {
          "kind": "range",
          "maximum": 1000,
          "minimum": 1
        }
      ],
      "description": "Step count aligned with every raw return.",
      "name": "episode_lengths",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 500,
          "minimum": 500
        }
      ],
      "description": "Environment termination flag for every rollout.",
      "name": "terminated",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 500,
          "minimum": 500
        }
      ],
      "description": "Time-limit truncation flag for every rollout.",
      "name": "truncated",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "range",
          "maximum": 500,
          "minimum": 500
        }
      ],
      "description": "Number of frozen rollouts per policy.",
      "name": "rollout_count",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "finite"
        }
      ],
      "description": "Median of the raw episodic returns.",
      "name": "median_return",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "finite"
        },
        {
          "kind": "range",
          "minimum": 0
        }
      ],
      "description": "Sample standard deviation of raw episodic returns.",
      "name": "return_std",
      "required": false
    },
    {
      "constraints": [
        {
          "kind": "finite"
        },
        {
          "kind": "range",
          "minimum": 0
        }
      ],
      "description": "Sample standard deviation divided by sqrt(500).",
      "name": "return_standard_error",
      "required": false
    }
  ],
  "dataset_id": "design-bench/hopper-controller",
  "default_split": "policies",
  "description": "Design-Bench structured PPO controller checkpoints evaluated by 500 frozen stochastic Hopper-v5 rollouts per policy.",
  "inputs": [
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 5126,
          "minimum": 5126
        },
        {
          "kind": "finite"
        }
      ],
      "description": "Float32 parameters of the 11-64-64-3 tanh Gaussian policy.",
      "name": "policy_weights"
    }
  ],
  "knowledge": {
    "continuous_control_and_gaussian_policies": {
      "description": "Continuous actions, diagonal Gaussian policies, log standard deviations, stochastic sampling, and bounded-action clipping.",
      "media_type": "text/markdown",
      "path": "knowledge/shared/continuous-control-and-gaussian-policies.md",
      "title": "Continuous Control and Gaussian Policies"
    },
    "feedforward_neural_policy_parameterization": {
      "description": "Dense neural policy equations, parameter blocks, tanh activation, flattening conventions, and hidden-unit symmetries.",
      "media_type": "text/markdown",
      "path": "knowledge/shared/feedforward-neural-policy-parameterization.md",
      "title": "Feedforward Neural Policy Parameterization"
    },
    "hopper_locomotion_and_mujoco_dynamics": {
      "description": "Hopper-v5 body, observations, torque actions, reward and episode semantics, contact-rich locomotion, and MuJoCo simulation.",
      "media_type": "text/markdown",
      "path": "knowledge/design-bench/hopper_controller/hopper-locomotion-and-mujoco-dynamics.md",
      "title": "Hopper Locomotion and MuJoCo Dynamics"
    },
    "proximal_policy_optimization_and_policy_checkpoints": {
      "description": "On-policy learning, the PPO clipped surrogate, actor and critic roles, stochastic optimization, and checkpoint interpretation.",
      "media_type": "text/markdown",
      "path": "knowledge/shared/proximal-policy-optimization-and-policy-checkpoints.md",
      "title": "Proximal Policy Optimization and Policy Checkpoints"
    },
    "reinforcement_learning_and_episodic_return": {
      "description": "Agent-environment interaction, policies, trajectories, episodic return, expected performance, termination, and truncation.",
      "media_type": "text/markdown",
      "path": "knowledge/shared/reinforcement-learning-and-episodic-return.md",
      "title": "Reinforcement Learning and Episodic Return"
    },
    "stochastic_rollout_evaluation_and_uncertainty": {
      "description": "Repeated policy returns, distribution summaries, sampling uncertainty, heteroscedasticity, and common random numbers.",
      "media_type": "text/markdown",
      "path": "knowledge/shared/stochastic-rollout-evaluation-and-uncertainty.md",
      "title": "Stochastic Rollout Evaluation and Uncertainty"
    }
  },
  "license": "MIT",
  "schema_version": 1,
  "source": [
    {
      "checksum": "sha256:da93f2033436ae5c2f31e584ae42b4b4d043bea29448d78eec6470e4d8ee71d7",
      "name": "Design-Bench Hopper Controller policies",
      "revision": "e52939588421b5433f6f2e9b359cf013c542bd89",
      "url": "https://storage.googleapis.com/design-bench/hopper_controller/hopper_controller-x-0.npy"
    },
    {
      "checksum": "sha256:8f04be4b81627e59bd83faf42e7aeaed3134dafc8f97d35c3fc887a649d93efd",
      "name": "Design-Bench Hopper Controller source targets",
      "revision": "e52939588421b5433f6f2e9b359cf013c542bd89",
      "url": "https://storage.googleapis.com/design-bench/hopper_controller/hopper_controller-y-0.npy"
    }
  ],
  "splits": [
    {
      "attributes": {
        "environment": "Hopper-v5",
        "rollouts_per_policy": 500,
        "target_aggregation": "arithmetic_mean"
      },
      "description": "All 3,200 source policies and frozen measurements.",
      "name": "policies",
      "num_rows": 3200
    }
  ],
  "targets": [
    {
      "constraints": [
        {
          "kind": "length",
          "maximum": 500,
          "minimum": 500
        },
        {
          "kind": "finite"
        }
      ],
      "description": "All 500 frozen stochastic episodic returns.",
      "name": "raw_returns"
    },
    {
      "constraints": [
        {
          "kind": "finite"
        }
      ],
      "description": "Arithmetic mean of the 500 episodic returns.",
      "name": "mean_return"
    }
  ],
  "version": "1.0.0"
}