anhtld commited on
Commit
f25eb87
·
verified ·
1 Parent(s): 0c7f146

ctt api upload 2026-07-02 workspace/runs/utility_energy_smoke/metrics.json

Browse files
workspace/runs/utility_energy_smoke/metrics.json ADDED
@@ -0,0 +1,594 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "data_hash": "90b431047a1f35cec095cbafa7de5bc0126cb0d0f4db3ff31843dca369977957",
3
+ "epochs": 1,
4
+ "loss_weights": {
5
+ "listwise_rank": 1.0,
6
+ "margin": 0.5,
7
+ "outcome": 1.0,
8
+ "pairwise_rank": 2.0,
9
+ "temperature": 1.0
10
+ },
11
+ "num_charts": 32,
12
+ "report_type": "utility_energy_train",
13
+ "rows": [
14
+ {
15
+ "chart_id": "ms-01e6eaaaf40875e0ef5a",
16
+ "mean_abs_outcome_error": 0.06975866109132767,
17
+ "mean_abs_utility_error": 0.8692922592163086,
18
+ "pairwise_accuracy": 0.5966386554621849,
19
+ "pairwise_causal_calibration_ece": 0.07345783732768185,
20
+ "seed": "1671659352",
21
+ "task_id": "StackCube-v1"
22
+ },
23
+ {
24
+ "chart_id": "ms-0350672b45f92c437c22",
25
+ "mean_abs_outcome_error": 0.03609073907136917,
26
+ "mean_abs_utility_error": 0.5264366865158081,
27
+ "pairwise_accuracy": 0.5833333333333334,
28
+ "pairwise_causal_calibration_ece": 0.07226753454553303,
29
+ "seed": "1753014417",
30
+ "task_id": "PushCube-v1"
31
+ },
32
+ {
33
+ "chart_id": "ms-15459afe7e32ce62fc1b",
34
+ "mean_abs_outcome_error": 0.04434176906943321,
35
+ "mean_abs_utility_error": 0.581527054309845,
36
+ "pairwise_accuracy": 0.6923076923076923,
37
+ "pairwise_causal_calibration_ece": 0.1555445426947485,
38
+ "seed": "1461069713",
39
+ "task_id": "PullCube-v1"
40
+ },
41
+ {
42
+ "chart_id": "ms-0a4d4909900fb0bc4a9a",
43
+ "mean_abs_outcome_error": 0.03463064506649971,
44
+ "mean_abs_utility_error": 0.5596855878829956,
45
+ "pairwise_accuracy": 0.5916666666666667,
46
+ "pairwise_causal_calibration_ece": 0.04994094741586674,
47
+ "seed": "1754115693",
48
+ "task_id": "StackCube-v1"
49
+ },
50
+ {
51
+ "chart_id": "ms-0949097d19e597fc7f63",
52
+ "mean_abs_outcome_error": 0.055287424474954605,
53
+ "mean_abs_utility_error": 0.7063317894935608,
54
+ "pairwise_accuracy": 0.7,
55
+ "pairwise_causal_calibration_ece": 0.1735221201437922,
56
+ "seed": "460988816",
57
+ "task_id": "StackCube-v1"
58
+ },
59
+ {
60
+ "chart_id": "ms-0285f5815c210ac00bec",
61
+ "mean_abs_outcome_error": 0.025612499564886093,
62
+ "mean_abs_utility_error": 0.4249221086502075,
63
+ "pairwise_accuracy": 0.6916666666666667,
64
+ "pairwise_causal_calibration_ece": 0.16814035014460327,
65
+ "seed": "451255434",
66
+ "task_id": "PullCube-v1"
67
+ },
68
+ {
69
+ "chart_id": "ms-003a2225c0892dcda683",
70
+ "mean_abs_outcome_error": 0.07978884130716324,
71
+ "mean_abs_utility_error": 0.7941392064094543,
72
+ "pairwise_accuracy": 0.7863247863247863,
73
+ "pairwise_causal_calibration_ece": 0.2573995192269217,
74
+ "seed": "585904130",
75
+ "task_id": "PullCube-v1"
76
+ },
77
+ {
78
+ "chart_id": "ms-0541a144afe57b5a7ae6",
79
+ "mean_abs_outcome_error": 0.05973448604345322,
80
+ "mean_abs_utility_error": 0.6622100472450256,
81
+ "pairwise_accuracy": 0.6916666666666667,
82
+ "pairwise_causal_calibration_ece": 0.17267163927357554,
83
+ "seed": "1982046304",
84
+ "task_id": "PickCube-v1"
85
+ },
86
+ {
87
+ "chart_id": "ms-0660d5f65df5fc4b6132",
88
+ "mean_abs_outcome_error": 0.04164355993270874,
89
+ "mean_abs_utility_error": 0.7494800090789795,
90
+ "pairwise_accuracy": 0.75,
91
+ "pairwise_causal_calibration_ece": 0.22331983511205478,
92
+ "seed": "692998894",
93
+ "task_id": "LiftPegUpright-v1"
94
+ },
95
+ {
96
+ "chart_id": "ms-1150c257eb77fdf40f55",
97
+ "mean_abs_outcome_error": 0.04095236584544182,
98
+ "mean_abs_utility_error": 0.5620254874229431,
99
+ "pairwise_accuracy": 0.525,
100
+ "pairwise_causal_calibration_ece": 0.0034121474393626894,
101
+ "seed": "848613296",
102
+ "task_id": "PullCube-v1"
103
+ },
104
+ {
105
+ "chart_id": "ms-07aa7d1f9c01e93d04bc",
106
+ "mean_abs_outcome_error": 0.06449826061725616,
107
+ "mean_abs_utility_error": 0.7696523666381836,
108
+ "pairwise_accuracy": 0.7692307692307693,
109
+ "pairwise_causal_calibration_ece": 0.2578268755856008,
110
+ "seed": "1054295834",
111
+ "task_id": "PushCube-v1"
112
+ },
113
+ {
114
+ "chart_id": "ms-0b1dd0e538b8df15a6e4",
115
+ "mean_abs_outcome_error": 0.04421722888946533,
116
+ "mean_abs_utility_error": 0.7255164980888367,
117
+ "pairwise_accuracy": 0.6416666666666667,
118
+ "pairwise_causal_calibration_ece": 0.1155098143008102,
119
+ "seed": "1231384138",
120
+ "task_id": "StackCube-v1"
121
+ },
122
+ {
123
+ "chart_id": "ms-13d8bae55d3c3710138f",
124
+ "mean_abs_outcome_error": 0.046550530940294266,
125
+ "mean_abs_utility_error": 0.7782812714576721,
126
+ "pairwise_accuracy": 0.7,
127
+ "pairwise_causal_calibration_ece": 0.17109290430604018,
128
+ "seed": "1148327469",
129
+ "task_id": "LiftPegUpright-v1"
130
+ },
131
+ {
132
+ "chart_id": "ms-0b93907b2a7261815f63",
133
+ "mean_abs_outcome_error": 0.029401155188679695,
134
+ "mean_abs_utility_error": 0.26239433884620667,
135
+ "pairwise_accuracy": 0.7,
136
+ "pairwise_causal_calibration_ece": 0.19213148845615236,
137
+ "seed": "626486813",
138
+ "task_id": "PullCube-v1"
139
+ },
140
+ {
141
+ "chart_id": "ms-016c5de39ed3c111889a",
142
+ "mean_abs_outcome_error": 0.040086567401885986,
143
+ "mean_abs_utility_error": 0.6989057660102844,
144
+ "pairwise_accuracy": 0.625,
145
+ "pairwise_causal_calibration_ece": 0.09649891842648839,
146
+ "seed": "2061635888",
147
+ "task_id": "StackCube-v1"
148
+ },
149
+ {
150
+ "chart_id": "ms-0b226017db7f07c5307e",
151
+ "mean_abs_outcome_error": 0.07301713526248932,
152
+ "mean_abs_utility_error": 0.8979111313819885,
153
+ "pairwise_accuracy": 0.7983193277310925,
154
+ "pairwise_causal_calibration_ece": 0.2748533906818609,
155
+ "seed": "2099816463",
156
+ "task_id": "StackCube-v1"
157
+ },
158
+ {
159
+ "chart_id": "ms-02311c7783bc67e188b4",
160
+ "mean_abs_outcome_error": 0.03562131151556969,
161
+ "mean_abs_utility_error": 0.5002692937850952,
162
+ "pairwise_accuracy": 0.725,
163
+ "pairwise_causal_calibration_ece": 0.2107236792507069,
164
+ "seed": "2015956957",
165
+ "task_id": "PullCube-v1"
166
+ },
167
+ {
168
+ "chart_id": "ms-1044e227d00afb01a5dd",
169
+ "mean_abs_outcome_error": 0.03491806238889694,
170
+ "mean_abs_utility_error": 0.5755202770233154,
171
+ "pairwise_accuracy": 0.6083333333333333,
172
+ "pairwise_causal_calibration_ece": 0.08908094932807686,
173
+ "seed": "1676419878",
174
+ "task_id": "PushCube-v1"
175
+ },
176
+ {
177
+ "chart_id": "ms-02fd0338e80ef19ceec1",
178
+ "mean_abs_outcome_error": 0.01940535381436348,
179
+ "mean_abs_utility_error": 0.5485053658485413,
180
+ "pairwise_accuracy": 0.48333333333333334,
181
+ "pairwise_causal_calibration_ece": 0.026630046978858235,
182
+ "seed": "585904130",
183
+ "task_id": "PullCube-v1"
184
+ },
185
+ {
186
+ "chart_id": "ms-07b479748a91a9c92b20",
187
+ "mean_abs_outcome_error": 0.039407987147569656,
188
+ "mean_abs_utility_error": 0.6002693772315979,
189
+ "pairwise_accuracy": 0.6,
190
+ "pairwise_causal_calibration_ece": 0.077571216040902,
191
+ "seed": "1867746099",
192
+ "task_id": "StackCube-v1"
193
+ },
194
+ {
195
+ "chart_id": "ms-0547201761e8e1873482",
196
+ "mean_abs_outcome_error": 0.07047047466039658,
197
+ "mean_abs_utility_error": 0.8302875757217407,
198
+ "pairwise_accuracy": 0.7226890756302521,
199
+ "pairwise_causal_calibration_ece": 0.19875563809574193,
200
+ "seed": "1369129094",
201
+ "task_id": "StackCube-v1"
202
+ },
203
+ {
204
+ "chart_id": "ms-1577a0922a9e53528571",
205
+ "mean_abs_outcome_error": 0.02661365270614624,
206
+ "mean_abs_utility_error": 0.42043066024780273,
207
+ "pairwise_accuracy": 0.7,
208
+ "pairwise_causal_calibration_ece": 0.16922042717698893,
209
+ "seed": "585904130",
210
+ "task_id": "PullCube-v1"
211
+ },
212
+ {
213
+ "chart_id": "ms-04eec0abab5f55620e4e",
214
+ "mean_abs_outcome_error": 0.06774069368839264,
215
+ "mean_abs_utility_error": 0.8256913423538208,
216
+ "pairwise_accuracy": 0.6638655462184874,
217
+ "pairwise_causal_calibration_ece": 0.140139987278719,
218
+ "seed": "1151410346",
219
+ "task_id": "StackCube-v1"
220
+ },
221
+ {
222
+ "chart_id": "ms-059d7962a5f998edf668",
223
+ "mean_abs_outcome_error": 0.06238825246691704,
224
+ "mean_abs_utility_error": 0.9378910660743713,
225
+ "pairwise_accuracy": 0.4957983193277311,
226
+ "pairwise_causal_calibration_ece": 0.019596244022319853,
227
+ "seed": "500109393",
228
+ "task_id": "LiftPegUpright-v1"
229
+ },
230
+ {
231
+ "chart_id": "ms-067cac2e4dbeddb12239",
232
+ "mean_abs_outcome_error": 0.050321225076913834,
233
+ "mean_abs_utility_error": 0.7952984571456909,
234
+ "pairwise_accuracy": 0.6386554621848739,
235
+ "pairwise_causal_calibration_ece": 0.12567522214134175,
236
+ "seed": "1461069713",
237
+ "task_id": "PullCube-v1"
238
+ },
239
+ {
240
+ "chart_id": "ms-076d67ac4d2de4e0c627",
241
+ "mean_abs_outcome_error": 0.05422291159629822,
242
+ "mean_abs_utility_error": 0.4873132109642029,
243
+ "pairwise_accuracy": 0.7333333333333333,
244
+ "pairwise_causal_calibration_ece": 0.21221080849778506,
245
+ "seed": "111895951",
246
+ "task_id": "PickCube-v1"
247
+ },
248
+ {
249
+ "chart_id": "ms-03df3e12e1d61aa9ebbf",
250
+ "mean_abs_outcome_error": 0.05661546066403389,
251
+ "mean_abs_utility_error": 0.7380750775337219,
252
+ "pairwise_accuracy": 0.6333333333333333,
253
+ "pairwise_causal_calibration_ece": 0.10660925539691046,
254
+ "seed": "1894022513",
255
+ "task_id": "StackCube-v1"
256
+ },
257
+ {
258
+ "chart_id": "ms-0093b06390890d181104",
259
+ "mean_abs_outcome_error": 0.02126285433769226,
260
+ "mean_abs_utility_error": 0.394942045211792,
261
+ "pairwise_accuracy": 0.55,
262
+ "pairwise_causal_calibration_ece": 0.040736455972501195,
263
+ "seed": "750448296",
264
+ "task_id": "PullCube-v1"
265
+ },
266
+ {
267
+ "chart_id": "ms-062addb9da32a0c2ea97",
268
+ "mean_abs_outcome_error": 0.04072529077529907,
269
+ "mean_abs_utility_error": 0.7450370788574219,
270
+ "pairwise_accuracy": 0.6666666666666666,
271
+ "pairwise_causal_calibration_ece": 0.14993840550534232,
272
+ "seed": "645859527",
273
+ "task_id": "LiftPegUpright-v1"
274
+ },
275
+ {
276
+ "chart_id": "ms-14c6516a0f22d20ef6b0",
277
+ "mean_abs_outcome_error": 0.0978042408823967,
278
+ "mean_abs_utility_error": 0.9380365610122681,
279
+ "pairwise_accuracy": 0.4818181818181818,
280
+ "pairwise_causal_calibration_ece": 0.02412205504980458,
281
+ "seed": "1824746139",
282
+ "task_id": "PushCube-v1"
283
+ },
284
+ {
285
+ "chart_id": "ms-1772611570ffaf72c98c",
286
+ "mean_abs_outcome_error": 0.016714444383978844,
287
+ "mean_abs_utility_error": 0.513826847076416,
288
+ "pairwise_accuracy": 0.375,
289
+ "pairwise_causal_calibration_ece": 0.13848124754891455,
290
+ "seed": "278008524",
291
+ "task_id": "PullCube-v1"
292
+ },
293
+ {
294
+ "chart_id": "ms-0e592951fa32139b6dc9",
295
+ "mean_abs_outcome_error": 0.017789587378501892,
296
+ "mean_abs_utility_error": 0.401470422744751,
297
+ "pairwise_accuracy": 0.5333333333333333,
298
+ "pairwise_causal_calibration_ece": 0.023225169087835784,
299
+ "seed": "565369392",
300
+ "task_id": "PullCube-v1"
301
+ }
302
+ ],
303
+ "split_hash": "867a94a1c67686c082765671a7bf8d74a9d633085e1b2619c83197c60bf58372",
304
+ "summary": {
305
+ "mean_abs_outcome_error": {
306
+ "macro_by_seed": {
307
+ "ci": {
308
+ "confidence": 0.95,
309
+ "high": 0.053616313799016775,
310
+ "low": 0.03968523808851324,
311
+ "mean": 0.047118285435369646,
312
+ "n": 29,
313
+ "num_samples": 200
314
+ },
315
+ "mean": 0.047118285435369646,
316
+ "per_seed": {
317
+ "1054295834": 0.06449826061725616,
318
+ "111895951": 0.05422291159629822,
319
+ "1148327469": 0.046550530940294266,
320
+ "1151410346": 0.06774069368839264,
321
+ "1231384138": 0.04421722888946533,
322
+ "1369129094": 0.07047047466039658,
323
+ "1461069713": 0.04733149707317352,
324
+ "1671659352": 0.06975866109132767,
325
+ "1676419878": 0.03491806238889694,
326
+ "1753014417": 0.03609073907136917,
327
+ "1754115693": 0.03463064506649971,
328
+ "1824746139": 0.0978042408823967,
329
+ "1867746099": 0.039407987147569656,
330
+ "1894022513": 0.05661546066403389,
331
+ "1982046304": 0.05973448604345322,
332
+ "2015956957": 0.03562131151556969,
333
+ "2061635888": 0.040086567401885986,
334
+ "2099816463": 0.07301713526248932,
335
+ "278008524": 0.016714444383978844,
336
+ "451255434": 0.025612499564886093,
337
+ "460988816": 0.055287424474954605,
338
+ "500109393": 0.06238825246691704,
339
+ "565369392": 0.017789587378501892,
340
+ "585904130": 0.041935949275890984,
341
+ "626486813": 0.029401155188679695,
342
+ "645859527": 0.04072529077529907,
343
+ "692998894": 0.04164355993270874,
344
+ "750448296": 0.02126285433769226,
345
+ "848613296": 0.04095236584544182
346
+ }
347
+ },
348
+ "macro_by_task": {
349
+ "ci": {
350
+ "confidence": 0.95,
351
+ "high": 0.05677616119384765,
352
+ "low": 0.04135237447917461,
353
+ "mean": 0.05044841652115186,
354
+ "n": 5,
355
+ "num_samples": 200
356
+ },
357
+ "mean": 0.05044841652115186,
358
+ "per_task": {
359
+ "LiftPegUpright-v1": 0.04782690852880478,
360
+ "PickCube-v1": 0.05697869881987572,
361
+ "PullCube-v1": 0.03398542168239752,
362
+ "PushCube-v1": 0.058327825739979744,
363
+ "StackCube-v1": 0.05512322783470154
364
+ }
365
+ },
366
+ "metric": "mean_abs_outcome_error",
367
+ "micro": {
368
+ "confidence": 0.95,
369
+ "high": 0.05253183585591614,
370
+ "low": 0.040026556758675724,
371
+ "mean": 0.0468010522890836,
372
+ "n": 32,
373
+ "num_samples": 200
374
+ },
375
+ "n": 32
376
+ },
377
+ "mean_abs_utility_error": {
378
+ "macro_by_seed": {
379
+ "ci": {
380
+ "confidence": 0.95,
381
+ "high": 0.7145269711812336,
382
+ "low": 0.5887628752609779,
383
+ "mean": 0.6537165525315821,
384
+ "n": 29,
385
+ "num_samples": 200
386
+ },
387
+ "mean": 0.6537165525315821,
388
+ "per_seed": {
389
+ "1054295834": 0.7696523666381836,
390
+ "111895951": 0.4873132109642029,
391
+ "1148327469": 0.7782812714576721,
392
+ "1151410346": 0.8256913423538208,
393
+ "1231384138": 0.7255164980888367,
394
+ "1369129094": 0.8302875757217407,
395
+ "1461069713": 0.6884127557277679,
396
+ "1671659352": 0.8692922592163086,
397
+ "1676419878": 0.5755202770233154,
398
+ "1753014417": 0.5264366865158081,
399
+ "1754115693": 0.5596855878829956,
400
+ "1824746139": 0.9380365610122681,
401
+ "1867746099": 0.6002693772315979,
402
+ "1894022513": 0.7380750775337219,
403
+ "1982046304": 0.6622100472450256,
404
+ "2015956957": 0.5002692937850952,
405
+ "2061635888": 0.6989057660102844,
406
+ "2099816463": 0.8979111313819885,
407
+ "278008524": 0.513826847076416,
408
+ "451255434": 0.4249221086502075,
409
+ "460988816": 0.7063317894935608,
410
+ "500109393": 0.9378910660743713,
411
+ "565369392": 0.401470422744751,
412
+ "585904130": 0.5876917441685995,
413
+ "626486813": 0.26239433884620667,
414
+ "645859527": 0.7450370788574219,
415
+ "692998894": 0.7494800090789795,
416
+ "750448296": 0.394942045211792,
417
+ "848613296": 0.5620254874229431
418
+ }
419
+ },
420
+ "macro_by_task": {
421
+ "ci": {
422
+ "confidence": 0.95,
423
+ "high": 0.739577716588974,
424
+ "low": 0.5739792183041572,
425
+ "mean": 0.6683376078804334,
426
+ "n": 5,
427
+ "num_samples": 200
428
+ },
429
+ "mean": 0.6683376078804334,
430
+ "per_task": {
431
+ "LiftPegUpright-v1": 0.8026723563671112,
432
+ "PickCube-v1": 0.5747616291046143,
433
+ "PullCube-v1": 0.5166459406415621,
434
+ "PushCube-v1": 0.7024114727973938,
435
+ "StackCube-v1": 0.7451966404914856
436
+ }
437
+ },
438
+ "metric": "mean_abs_utility_error",
439
+ "micro": {
440
+ "confidence": 0.95,
441
+ "high": 0.7037879945710301,
442
+ "low": 0.5897194379940629,
443
+ "mean": 0.6506742583587766,
444
+ "n": 32,
445
+ "num_samples": 200
446
+ },
447
+ "n": 32
448
+ },
449
+ "pairwise_accuracy": {
450
+ "macro_by_seed": {
451
+ "ci": {
452
+ "confidence": 0.95,
453
+ "high": 0.6674788512516707,
454
+ "low": 0.6047838233436612,
455
+ "mean": 0.637082557214403,
456
+ "n": 29,
457
+ "num_samples": 200
458
+ },
459
+ "mean": 0.637082557214403,
460
+ "per_seed": {
461
+ "1054295834": 0.7692307692307693,
462
+ "111895951": 0.7333333333333333,
463
+ "1148327469": 0.7,
464
+ "1151410346": 0.6638655462184874,
465
+ "1231384138": 0.6416666666666667,
466
+ "1369129094": 0.7226890756302521,
467
+ "1461069713": 0.6654815772462831,
468
+ "1671659352": 0.5966386554621849,
469
+ "1676419878": 0.6083333333333333,
470
+ "1753014417": 0.5833333333333334,
471
+ "1754115693": 0.5916666666666667,
472
+ "1824746139": 0.4818181818181818,
473
+ "1867746099": 0.6,
474
+ "1894022513": 0.6333333333333333,
475
+ "1982046304": 0.6916666666666667,
476
+ "2015956957": 0.725,
477
+ "2061635888": 0.625,
478
+ "2099816463": 0.7983193277310925,
479
+ "278008524": 0.375,
480
+ "451255434": 0.6916666666666667,
481
+ "460988816": 0.7,
482
+ "500109393": 0.4957983193277311,
483
+ "565369392": 0.5333333333333333,
484
+ "585904130": 0.6565527065527065,
485
+ "626486813": 0.7,
486
+ "645859527": 0.6666666666666666,
487
+ "692998894": 0.75,
488
+ "750448296": 0.55,
489
+ "848613296": 0.525
490
+ }
491
+ },
492
+ "macro_by_task": {
493
+ "ci": {
494
+ "confidence": 0.95,
495
+ "high": 0.6823072733366851,
496
+ "low": 0.6215821868763045,
497
+ "mean": 0.650066303522186,
498
+ "n": 5,
499
+ "num_samples": 200
500
+ },
501
+ "mean": 0.650066303522186,
502
+ "per_task": {
503
+ "LiftPegUpright-v1": 0.6531162464985995,
504
+ "PickCube-v1": 0.7124999999999999,
505
+ "PullCube-v1": 0.616718439512557,
506
+ "PushCube-v1": 0.6106789044289045,
507
+ "StackCube-v1": 0.6573179271708683
508
+ }
509
+ },
510
+ "metric": "pairwise_accuracy",
511
+ "micro": {
512
+ "confidence": 0.95,
513
+ "high": 0.6738964517916723,
514
+ "low": 0.6058646316918375,
515
+ "mean": 0.6391869109240432,
516
+ "n": 32,
517
+ "num_samples": 200
518
+ },
519
+ "n": 32
520
+ },
521
+ "pairwise_causal_calibration_ece": {
522
+ "macro_by_seed": {
523
+ "ci": {
524
+ "confidence": 0.95,
525
+ "high": 0.15511241553971314,
526
+ "low": 0.1047316794208077,
527
+ "mean": 0.1299148319924351,
528
+ "n": 29,
529
+ "num_samples": 200
530
+ },
531
+ "mean": 0.1299148319924351,
532
+ "per_seed": {
533
+ "1054295834": 0.2578268755856008,
534
+ "111895951": 0.21221080849778506,
535
+ "1148327469": 0.17109290430604018,
536
+ "1151410346": 0.140139987278719,
537
+ "1231384138": 0.1155098143008102,
538
+ "1369129094": 0.19875563809574193,
539
+ "1461069713": 0.14060988241804512,
540
+ "1671659352": 0.07345783732768185,
541
+ "1676419878": 0.08908094932807686,
542
+ "1753014417": 0.07226753454553303,
543
+ "1754115693": 0.04994094741586674,
544
+ "1824746139": 0.02412205504980458,
545
+ "1867746099": 0.077571216040902,
546
+ "1894022513": 0.10660925539691046,
547
+ "1982046304": 0.17267163927357554,
548
+ "2015956957": 0.2107236792507069,
549
+ "2061635888": 0.09649891842648839,
550
+ "2099816463": 0.2748533906818609,
551
+ "278008524": 0.13848124754891455,
552
+ "451255434": 0.16814035014460327,
553
+ "460988816": 0.1735221201437922,
554
+ "500109393": 0.019596244022319853,
555
+ "565369392": 0.023225169087835784,
556
+ "585904130": 0.15108333112758962,
557
+ "626486813": 0.19213148845615236,
558
+ "645859527": 0.14993840550534232,
559
+ "692998894": 0.22331983511205478,
560
+ "750448296": 0.040736455972501195,
561
+ "848613296": 0.0034121474393626894
562
+ }
563
+ },
564
+ "macro_by_task": {
565
+ "ci": {
566
+ "confidence": 0.95,
567
+ "high": 0.1667905884355659,
568
+ "low": 0.12082916412581564,
569
+ "mean": 0.14017633905403243,
570
+ "n": 5,
571
+ "num_samples": 200
572
+ },
573
+ "mean": 0.14017633905403243,
574
+ "per_task": {
575
+ "LiftPegUpright-v1": 0.14098684723643928,
576
+ "PickCube-v1": 0.1924412238856803,
577
+ "PullCube-v1": 0.12594335800991133,
578
+ "PushCube-v1": 0.11082435362725382,
579
+ "StackCube-v1": 0.13068591251087736
580
+ }
581
+ },
582
+ "metric": "pairwise_causal_calibration_ece",
583
+ "micro": {
584
+ "confidence": 0.95,
585
+ "high": 0.1612145633668989,
586
+ "low": 0.10937866048886753,
587
+ "mean": 0.1315720835141826,
588
+ "n": 32,
589
+ "num_samples": 200
590
+ },
591
+ "n": 32
592
+ }
593
+ }
594
+ }