thanhkt's picture
Upload eval from run_all.sh (part 2)
99bd6ac verified
Raw History Blame Contribute Delete
21.1 kB
{
"per_task": [
{
"task_group": "nexarm",
"task_id": 0,
"metrics": {
"sum_rewards": [
19.99974172941686,
19.999361957517646,
0.09529052628987866,
19.999188637931343,
19.999270608643606,
19.9992777478245,
19.999105284953263,
19.999418590257783,
20.000916084683926,
19.99934790335388,
19.999441898509225,
19.99855993369729,
20.000135085231243,
20.00036255548894,
19.99929530569687,
19.99901467169276,
19.99965857824038,
19.998859509067078,
19.999555595867243,
19.9994569103765,
19.99985531412613,
19.998779777987995,
0.10474769178772037,
19.9991507777862,
19.999665855566953,
19.99976391548153,
19.99937519088873,
19.999789464676386,
19.99892399145604,
20.00058700809152,
19.99888170004406,
0.038407886172903805,
19.999002096083213,
0.3174886601296423,
20.000375268562795,
19.99882321007272,
19.999868311628127,
20.000463377952194,
20.000718933888276,
19.998874396568954,
19.99966454912113,
19.999821159472905,
19.99832065039298,
20.000663036384665,
19.99991149341564,
19.998641353576787,
19.999142178288174,
19.999830717745947,
19.999877647273827,
19.99981914113524
],
"max_rewards": [
10.0,
10.0,
1.114471630191873,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
0.0249885570106535,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
0.035302288875288856,
10.0,
1.3545529658928146,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0,
10.0
],
"successes": [
true,
true,
false,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
false,
true,
true,
true,
true,
true,
true,
true,
true,
false,
true,
false,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true
],
"video_paths": [
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_9.mp4"
],
"predicted_video_paths": [],
"stages": [
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": false,
"released_middle": false,
"picked_top": false,
"nested_top": false,
"released_top": false
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": false,
"nested_middle": false,
"released_middle": false,
"picked_top": false,
"nested_top": false,
"released_top": false
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": false,
"nested_middle": false,
"released_middle": false,
"picked_top": false,
"nested_top": false,
"released_top": false
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": false,
"nested_middle": false,
"released_middle": false,
"picked_top": false,
"nested_top": false,
"released_top": false
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
},
{
"picked_middle": true,
"nested_middle": true,
"released_middle": true,
"picked_top": true,
"nested_top": true,
"released_top": true
}
],
"pc_picked_middle": 94.0,
"pc_nested_middle": 92.0,
"pc_released_middle": 92.0,
"pc_picked_top": 92.0,
"pc_nested_top": 92.0,
"pc_released_top": 92.0
},
"n_episodes": 50,
"n_success": 46,
"pc_success": 92.0,
"pc_success_ci95": [
81.16175308165717,
96.84504859114071
]
}
],
"per_group": {
"nexarm": {
"avg_sum_reward": 18.410689877409993,
"avg_max_reward": 9.250586308839413,
"pc_success": 92.0,
"n_episodes": 50,
"n_success": 46,
"pc_success_ci95": [
81.16175308165717,
96.84504859114071
],
"video_paths": [
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_9.mp4"
],
"predicted_video_paths": [],
"pc_picked_middle": 94.0,
"pc_nested_middle": 92.0,
"pc_released_middle": 92.0,
"pc_picked_top": 92.0,
"pc_nested_top": 92.0,
"pc_released_top": 92.0
}
},
"overall": {
"avg_sum_reward": 18.410689877409993,
"avg_max_reward": 9.250586308839413,
"pc_success": 92.0,
"n_episodes": 50,
"n_success": 46,
"pc_success_ci95": [
81.16175308165717,
96.84504859114071
],
"eval_s": 316.0980794429779,
"eval_ep_s": 6.321961598396301,
"video_paths": [
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_latency/videos/nexarm_0/eval_episode_9.mp4"
],
"predicted_video_paths": [],
"pc_picked_middle": 94.0,
"pc_nested_middle": 92.0,
"pc_released_middle": 92.0,
"pc_picked_top": 92.0,
"pc_nested_top": 92.0,
"pc_released_top": 92.0
}
}