thanhkt's picture
Upload eval from run_all.sh (part 2)
99bd6ac verified
Raw History Blame Contribute Delete
21.2 kB
{
"per_task": [
{
"task_group": "nexarm",
"task_id": 0,
"metrics": {
"sum_rewards": [
19.999200674232068,
19.999369886234135,
19.99926099578967,
20.000081126184032,
19.99928110256385,
19.99930992585632,
19.9990972910451,
19.99936961735156,
20.000909302816158,
19.999347385874817,
19.999412300597417,
19.998563966808543,
20.00012852565214,
20.000383536146487,
19.999424195025775,
19.998971468835904,
19.999659755674088,
19.998857507902404,
19.999542603424516,
19.998809129351322,
19.999971749622198,
19.998493250340516,
19.998620713599518,
19.999150746526304,
19.999665839034755,
19.99974619931265,
19.999378407420647,
19.999694441253563,
19.998924166880442,
20.000592346032054,
19.998836058577407,
20.00117212580638,
19.999001967110708,
19.999954358065626,
20.000256182435216,
19.998585489611024,
19.999788067011526,
20.00031342182561,
20.000649256451027,
19.998868264937038,
19.999660102775426,
19.9997992209001,
19.9998926826476,
20.000744187708868,
20.001029153437166,
19.998615206851703,
19.999142178288174,
19.999822913215162,
20.00002250295301,
19.999815838005844
],
"max_rewards": [
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,
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,
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,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
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_none_every12/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/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": 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": 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": 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": 100.0,
"pc_nested_middle": 100.0,
"pc_released_middle": 100.0,
"pc_picked_top": 100.0,
"pc_nested_top": 100.0,
"pc_released_top": 100.0
},
"n_episodes": 50,
"n_success": 50,
"pc_success": 100.0,
"pc_success_ci95": [
92.86524008666413,
100.0
]
}
],
"per_group": {
"nexarm": {
"avg_sum_reward": 19.99958374672007,
"avg_max_reward": 10.0,
"pc_success": 100.0,
"n_episodes": 50,
"n_success": 50,
"pc_success_ci95": [
92.86524008666413,
100.0
],
"video_paths": [
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_9.mp4"
],
"predicted_video_paths": [],
"pc_picked_middle": 100.0,
"pc_nested_middle": 100.0,
"pc_released_middle": 100.0,
"pc_picked_top": 100.0,
"pc_nested_top": 100.0,
"pc_released_top": 100.0
}
},
"overall": {
"avg_sum_reward": 19.99958374672007,
"avg_max_reward": 10.0,
"pc_success": 100.0,
"n_episodes": 50,
"n_success": 50,
"pc_success_ci95": [
92.86524008666413,
100.0
],
"eval_s": 208.2226161956787,
"eval_ep_s": 4.164452333450317,
"video_paths": [
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_0.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_1.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_2.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_3.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_4.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_5.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_6.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_7.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_8.mp4",
"/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none_every12/videos/nexarm_0/eval_episode_9.mp4"
],
"predicted_video_paths": [],
"pc_picked_middle": 100.0,
"pc_nested_middle": 100.0,
"pc_released_middle": 100.0,
"pc_picked_top": 100.0,
"pc_nested_top": 100.0,
"pc_released_top": 100.0
}
}