jalf/TIR_PROJ
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Run v3

Browse Source
This commit is contained in:
Johnny Fernandes
2026-04-26 16:01:30 +00:00
parent e2883212c5
commit 27fe6d1bf5
36 changed files with 479 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
training/runs/v3.log
+242
View File
@@ -0,0 +1,242 @@
Config loaded from config.json
Config: {'W_PER_SHEEP': 2.0, 'W_ALIGN': 0.05, 'W_PEN_BONUS': 10.0, 'W_COMPLETE': 100.0, 'W_STEP_COST': 0.02, 'W_SOUTH': 0.01, 'W_COMPACT': 0.0, 'W_WALL_TOUCH': 0.0, 'WALL_TOUCH_BUFFER': 0.4, 'ALIGN_SHAPE': 'standoff', 'ALIGN_GATED': True, 'ENTRY_AWARE': True, 'ent_coef': 0.02}
Run dir: runs/v3
Curriculum: 1 → 10 sheep, 1,500,000 steps/stage
[Stage n_sheep=1] training 1,500,000 steps
... [1 sheep | 100,000 steps | ret(last 24)=-47.74 win_sr=12% cum_sr=12%]
... [1 sheep | 200,000 steps | ret(last 50)=-40.77 win_sr=14% cum_sr=16%]
... [1 sheep | 300,000 steps | ret(last 50)=-36.39 win_sr=16% cum_sr=16%]
... [1 sheep | 400,000 steps | ret(last 50)=-40.04 win_sr=14% cum_sr=15%]
... [1 sheep | 500,000 steps | ret(last 50)=+7.09 win_sr=80% cum_sr=36%]
... [1 sheep | 600,000 steps | ret(last 50)=+15.87 win_sr=100% cum_sr=71%]
... [1 sheep | 700,000 steps | ret(last 50)=+14.78 win_sr=100% cum_sr=84%]
... [1 sheep | 800,000 steps | ret(last 50)=+14.04 win_sr=100% cum_sr=90%]
... [1 sheep | 900,000 steps | ret(last 50)=+14.08 win_sr=100% cum_sr=92%]
... [1 sheep | 1,000,000 steps | ret(last 50)=+13.33 win_sr=100% cum_sr=94%]
... [1 sheep | 1,100,000 steps | ret(last 50)=+13.99 win_sr=100% cum_sr=95%]
... [1 sheep | 1,200,000 steps | ret(last 50)=+13.38 win_sr=100% cum_sr=96%]
... [1 sheep | 1,300,000 steps | ret(last 50)=+13.18 win_sr=100% cum_sr=96%]
... [1 sheep | 1,400,000 steps | ret(last 50)=+13.53 win_sr=100% cum_sr=97%]
... [1 sheep | 1,500,000 steps | ret(last 50)=+13.46 win_sr=100% cum_sr=97%]
[Stage n_sheep=1] evaluating 30 eps
[Stage n_sheep=1] sr=100% mean_len=264 mean_min_pen=3.7m mean_act=0.45
failure modes: SUCCESS=30
reward/step: progress=+0.1156 alignment=+0.0001 south=-0.0005 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0378 step_cost=-0.0200 complete=+0.3784
[Stage n_sheep=2] training 1,500,000 steps
... [2 sheep | 1,507,336 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [2 sheep | 1,607,336 steps | ret(last 35)=-3.04 win_sr=49% cum_sr=49%]
... [2 sheep | 1,707,336 steps | ret(last 50)=-11.13 win_sr=20% cum_sr=33%]
... [2 sheep | 1,807,336 steps | ret(last 50)=-11.83 win_sr=18% cum_sr=31%]
... [2 sheep | 1,907,336 steps | ret(last 50)=-8.76 win_sr=30% cum_sr=31%]
... [2 sheep | 2,007,336 steps | ret(last 50)=-8.95 win_sr=30% cum_sr=30%]
... [2 sheep | 2,107,336 steps | ret(last 50)=-9.06 win_sr=32% cum_sr=30%]
... [2 sheep | 2,207,336 steps | ret(last 50)=-9.48 win_sr=32% cum_sr=30%]
... [2 sheep | 2,307,336 steps | ret(last 50)=-1.70 win_sr=44% cum_sr=33%]
... [2 sheep | 2,407,336 steps | ret(last 50)=+5.02 win_sr=64% cum_sr=38%]
... [2 sheep | 2,507,336 steps | ret(last 50)=+13.32 win_sr=88% cum_sr=46%]
... [2 sheep | 2,607,336 steps | ret(last 50)=+12.15 win_sr=90% cum_sr=54%]
... [2 sheep | 2,707,336 steps | ret(last 50)=+17.13 win_sr=98% cum_sr=63%]
... [2 sheep | 2,807,336 steps | ret(last 50)=+18.81 win_sr=98% cum_sr=69%]
... [2 sheep | 2,907,336 steps | ret(last 50)=+16.23 win_sr=92% cum_sr=73%]
... [2 sheep | 3,007,336 steps | ret(last 50)=+18.83 win_sr=100% cum_sr=76%]
[Stage n_sheep=2] evaluating 30 eps
[Stage n_sheep=2] sr=77% mean_len=1398 mean_min_pen=3.3m mean_act=0.97
failure modes: SUCCESS=23 PARTIAL_1of2=6 COMPACT_CANT_DRIVE=1
reward/step: progress=+0.0401 alignment=+0.0045 south=-0.0039 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0126 step_cost=-0.0200 complete=+0.0549
[Stage n_sheep=3] training 1,500,000 steps
... [3 sheep | 3,014,664 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [3 sheep | 3,114,664 steps | ret(last 50)=+13.79 win_sr=82% cum_sr=84%]
... [3 sheep | 3,214,664 steps | ret(last 50)=+21.64 win_sr=96% cum_sr=88%]
... [3 sheep | 3,314,664 steps | ret(last 50)=+23.45 win_sr=98% cum_sr=92%]
... [3 sheep | 3,414,664 steps | ret(last 50)=+22.18 win_sr=98% cum_sr=94%]
... [3 sheep | 3,514,664 steps | ret(last 50)=+24.83 win_sr=100% cum_sr=96%]
... [3 sheep | 3,614,664 steps | ret(last 50)=+19.77 win_sr=94% cum_sr=96%]
... [3 sheep | 3,714,664 steps | ret(last 50)=+25.53 win_sr=100% cum_sr=96%]
... [3 sheep | 3,814,664 steps | ret(last 50)=+25.24 win_sr=100% cum_sr=97%]
... [3 sheep | 3,914,664 steps | ret(last 50)=+24.43 win_sr=100% cum_sr=97%]
... [3 sheep | 4,014,664 steps | ret(last 50)=+24.59 win_sr=100% cum_sr=97%]
... [3 sheep | 4,114,664 steps | ret(last 50)=+22.18 win_sr=98% cum_sr=98%]
... [3 sheep | 4,214,664 steps | ret(last 50)=+23.11 win_sr=96% cum_sr=97%]
... [3 sheep | 4,314,664 steps | ret(last 50)=+23.06 win_sr=98% cum_sr=97%]
... [3 sheep | 4,414,664 steps | ret(last 50)=+23.35 win_sr=100% cum_sr=97%]
... [3 sheep | 4,514,664 steps | ret(last 50)=+22.50 win_sr=100% cum_sr=98%]
[Stage n_sheep=3] evaluating 30 eps
[Stage n_sheep=3] sr=97% mean_len=1095 mean_min_pen=2.5m mean_act=0.95
failure modes: SUCCESS=29 COMPACT_CANT_DRIVE=1
reward/step: progress=+0.0821 alignment=+0.0113 south=-0.0087 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0265 step_cost=-0.0200 complete=+0.0883
[Stage n_sheep=4] training 1,500,000 steps
... [4 sheep | 4,521,992 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [4 sheep | 4,621,992 steps | ret(last 50)=+22.17 win_sr=92% cum_sr=94%]
... [4 sheep | 4,721,992 steps | ret(last 50)=+25.81 win_sr=94% cum_sr=93%]
... [4 sheep | 4,821,992 steps | ret(last 50)=+21.80 win_sr=90% cum_sr=93%]
... [4 sheep | 4,921,992 steps | ret(last 50)=+26.38 win_sr=98% cum_sr=94%]
... [4 sheep | 5,021,992 steps | ret(last 50)=+26.65 win_sr=98% cum_sr=95%]
... [4 sheep | 5,121,992 steps | ret(last 50)=+26.07 win_sr=98% cum_sr=95%]
... [4 sheep | 5,221,992 steps | ret(last 50)=+27.08 win_sr=98% cum_sr=96%]
... [4 sheep | 5,321,992 steps | ret(last 50)=+27.87 win_sr=100% cum_sr=96%]
... [4 sheep | 5,421,992 steps | ret(last 50)=+27.53 win_sr=100% cum_sr=97%]
... [4 sheep | 5,521,992 steps | ret(last 50)=+25.91 win_sr=100% cum_sr=97%]
... [4 sheep | 5,621,992 steps | ret(last 50)=+27.75 win_sr=100% cum_sr=97%]
... [4 sheep | 5,721,992 steps | ret(last 50)=+25.63 win_sr=100% cum_sr=97%]
... [4 sheep | 5,821,992 steps | ret(last 50)=+24.43 win_sr=98% cum_sr=97%]
... [4 sheep | 5,921,992 steps | ret(last 50)=+22.52 win_sr=94% cum_sr=97%]
... [4 sheep | 6,021,992 steps | ret(last 50)=+27.28 win_sr=100% cum_sr=98%]
[Stage n_sheep=4] evaluating 30 eps
[Stage n_sheep=4] sr=57% mean_len=2572 mean_min_pen=2.2m mean_act=1.28
failure modes: SUCCESS=17 PARTIAL_1of4=6 PARTIAL_2of4=5 DROVE_NO_SHEEP=1 NEVER_COMPACT=1
reward/step: progress=+0.0455 alignment=+0.0040 south=-0.0454 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0109 step_cost=-0.0200 complete=+0.0220
[Stage n_sheep=5] training 1,500,000 steps
... [5 sheep | 6,029,320 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [5 sheep | 6,129,320 steps | ret(last 50)=+28.06 win_sr=96% cum_sr=96%]
... [5 sheep | 6,229,320 steps | ret(last 50)=+31.40 win_sr=98% cum_sr=96%]
... [5 sheep | 6,329,320 steps | ret(last 50)=+27.81 win_sr=96% cum_sr=96%]
... [5 sheep | 6,429,320 steps | ret(last 50)=+22.08 win_sr=88% cum_sr=95%]
... [5 sheep | 6,529,320 steps | ret(last 50)=+26.99 win_sr=94% cum_sr=95%]
... [5 sheep | 6,629,320 steps | ret(last 50)=+21.24 win_sr=86% cum_sr=93%]
... [5 sheep | 6,729,320 steps | ret(last 50)=+24.58 win_sr=94% cum_sr=93%]
... [5 sheep | 6,829,320 steps | ret(last 50)=+29.66 win_sr=96% cum_sr=93%]
... [5 sheep | 6,929,320 steps | ret(last 50)=+27.53 win_sr=96% cum_sr=93%]
... [5 sheep | 7,029,320 steps | ret(last 50)=+28.99 win_sr=100% cum_sr=94%]
... [5 sheep | 7,129,320 steps | ret(last 50)=+27.59 win_sr=98% cum_sr=94%]
... [5 sheep | 7,229,320 steps | ret(last 50)=+30.79 win_sr=100% cum_sr=95%]
... [5 sheep | 7,329,320 steps | ret(last 50)=+30.56 win_sr=98% cum_sr=95%]
... [5 sheep | 7,429,320 steps | ret(last 50)=+31.55 win_sr=100% cum_sr=95%]
... [5 sheep | 7,529,320 steps | ret(last 50)=+29.95 win_sr=100% cum_sr=96%]
[Stage n_sheep=5] evaluating 30 eps
[Stage n_sheep=5] sr=0% mean_len=4000 mean_min_pen=1.7m mean_act=1.36
failure modes: PARTIAL_4of5=17 PARTIAL_1of5=9 PARTIAL_3of5=2 PARTIAL_2of5=2
reward/step: progress=+0.0396 alignment=+0.0034 south=-0.0393 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0073 step_cost=-0.0200 complete=+0.0000
[Stage n_sheep=6] training 1,500,000 steps
... [6 sheep | 7,536,648 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [6 sheep | 7,636,648 steps | ret(last 50)=+34.50 win_sr=100% cum_sr=100%]
... [6 sheep | 7,736,648 steps | ret(last 50)=+31.01 win_sr=100% cum_sr=100%]
... [6 sheep | 7,836,648 steps | ret(last 50)=+33.27 win_sr=100% cum_sr=100%]
... [6 sheep | 7,936,648 steps | ret(last 50)=+34.81 win_sr=100% cum_sr=100%]
... [6 sheep | 8,036,648 steps | ret(last 50)=+32.69 win_sr=100% cum_sr=100%]
... [6 sheep | 8,136,648 steps | ret(last 50)=+31.36 win_sr=96% cum_sr=99%]
... [6 sheep | 8,236,648 steps | ret(last 50)=+33.71 win_sr=100% cum_sr=99%]
... [6 sheep | 8,336,648 steps | ret(last 50)=+34.71 win_sr=100% cum_sr=99%]
... [6 sheep | 8,436,648 steps | ret(last 50)=+31.89 win_sr=96% cum_sr=99%]
... [6 sheep | 8,536,648 steps | ret(last 50)=+35.63 win_sr=100% cum_sr=99%]
... [6 sheep | 8,636,648 steps | ret(last 50)=+35.92 win_sr=100% cum_sr=99%]
... [6 sheep | 8,736,648 steps | ret(last 50)=+33.70 win_sr=100% cum_sr=99%]
... [6 sheep | 8,836,648 steps | ret(last 50)=+33.46 win_sr=100% cum_sr=99%]
... [6 sheep | 8,936,648 steps | ret(last 50)=+35.12 win_sr=100% cum_sr=99%]
... [6 sheep | 9,036,648 steps | ret(last 50)=+34.21 win_sr=100% cum_sr=100%]
[Stage n_sheep=6] evaluating 30 eps
[Stage n_sheep=6] sr=37% mean_len=3137 mean_min_pen=1.8m mean_act=1.37
failure modes: PARTIAL_4of6=14 SUCCESS=11 PARTIAL_3of6=5
reward/step: progress=+0.0654 alignment=+0.0085 south=-0.0392 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0146 step_cost=-0.0200 complete=+0.0117
[Stage n_sheep=7] training 1,500,000 steps
... [7 sheep | 9,043,976 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [7 sheep | 9,143,976 steps | ret(last 50)=+36.14 win_sr=100% cum_sr=100%]
... [7 sheep | 9,243,976 steps | ret(last 50)=+33.77 win_sr=98% cum_sr=99%]
... [7 sheep | 9,343,976 steps | ret(last 50)=+37.14 win_sr=100% cum_sr=100%]
... [7 sheep | 9,443,976 steps | ret(last 50)=+39.90 win_sr=100% cum_sr=100%]
... [7 sheep | 9,543,976 steps | ret(last 50)=+37.52 win_sr=100% cum_sr=100%]
... [7 sheep | 9,643,976 steps | ret(last 50)=+37.31 win_sr=100% cum_sr=100%]
... [7 sheep | 9,743,976 steps | ret(last 50)=+36.24 win_sr=100% cum_sr=100%]
... [7 sheep | 9,843,976 steps | ret(last 50)=+39.67 win_sr=100% cum_sr=100%]
... [7 sheep | 9,943,976 steps | ret(last 50)=+39.12 win_sr=100% cum_sr=100%]
... [7 sheep | 10,043,976 steps | ret(last 50)=+37.82 win_sr=100% cum_sr=100%]
... [7 sheep | 10,143,976 steps | ret(last 50)=+37.38 win_sr=100% cum_sr=100%]
... [7 sheep | 10,243,976 steps | ret(last 50)=+37.47 win_sr=98% cum_sr=100%]
... [7 sheep | 10,343,976 steps | ret(last 50)=+36.04 win_sr=98% cum_sr=99%]
... [7 sheep | 10,443,976 steps | ret(last 50)=+31.71 win_sr=98% cum_sr=99%]
... [7 sheep | 10,543,976 steps | ret(last 50)=+32.50 win_sr=96% cum_sr=99%]
[Stage n_sheep=7] evaluating 30 eps
[Stage n_sheep=7] sr=0% mean_len=4000 mean_min_pen=1.8m mean_act=1.38
failure modes: PARTIAL_5of7=18 PARTIAL_6of7=7 PARTIAL_3of7=3 PARTIAL_4of7=2
reward/step: progress=+0.0533 alignment=+0.0069 south=-0.0356 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0124 step_cost=-0.0200 complete=+0.0000
[Stage n_sheep=8] training 1,500,000 steps
... [8 sheep | 10,551,304 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [8 sheep | 10,651,304 steps | ret(last 50)=+36.01 win_sr=96% cum_sr=96%]
... [8 sheep | 10,751,304 steps | ret(last 50)=+37.97 win_sr=96% cum_sr=96%]
... [8 sheep | 10,851,304 steps | ret(last 50)=+39.12 win_sr=100% cum_sr=98%]
... [8 sheep | 10,951,304 steps | ret(last 50)=+36.54 win_sr=96% cum_sr=97%]
... [8 sheep | 11,051,304 steps | ret(last 50)=+40.58 win_sr=100% cum_sr=98%]
... [8 sheep | 11,151,304 steps | ret(last 50)=+39.00 win_sr=98% cum_sr=98%]
... [8 sheep | 11,251,304 steps | ret(last 50)=+38.54 win_sr=98% cum_sr=98%]
... [8 sheep | 11,351,304 steps | ret(last 50)=+39.29 win_sr=100% cum_sr=98%]
... [8 sheep | 11,451,304 steps | ret(last 50)=+38.36 win_sr=100% cum_sr=98%]
... [8 sheep | 11,551,304 steps | ret(last 50)=+40.04 win_sr=100% cum_sr=98%]
... [8 sheep | 11,651,304 steps | ret(last 50)=+37.92 win_sr=100% cum_sr=99%]
... [8 sheep | 11,751,304 steps | ret(last 50)=+40.01 win_sr=98% cum_sr=99%]
... [8 sheep | 11,851,304 steps | ret(last 50)=+39.06 win_sr=100% cum_sr=99%]
... [8 sheep | 11,951,304 steps | ret(last 50)=+41.39 win_sr=100% cum_sr=99%]
... [8 sheep | 12,051,304 steps | ret(last 50)=+40.05 win_sr=100% cum_sr=99%]
[Stage n_sheep=8] evaluating 30 eps
[Stage n_sheep=8] sr=60% mean_len=2472 mean_min_pen=1.6m mean_act=1.39
failure modes: SUCCESS=18 PARTIAL_6of8=9 PARTIAL_4of8=3
reward/step: progress=+0.0956 alignment=+0.0106 south=-0.0508 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0283 step_cost=-0.0200 complete=+0.0243
[Stage n_sheep=9] training 1,500,000 steps
... [9 sheep | 12,058,632 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [9 sheep | 12,158,632 steps | ret(last 50)=+41.35 win_sr=98% cum_sr=98%]
... [9 sheep | 12,258,632 steps | ret(last 50)=+41.63 win_sr=100% cum_sr=99%]
... [9 sheep | 12,358,632 steps | ret(last 50)=+41.85 win_sr=100% cum_sr=99%]
... [9 sheep | 12,458,632 steps | ret(last 50)=+42.49 win_sr=100% cum_sr=100%]
... [9 sheep | 12,558,632 steps | ret(last 50)=+40.87 win_sr=100% cum_sr=100%]
... [9 sheep | 12,658,632 steps | ret(last 50)=+39.09 win_sr=100% cum_sr=100%]
... [9 sheep | 12,758,632 steps | ret(last 50)=+42.23 win_sr=100% cum_sr=100%]
... [9 sheep | 12,858,632 steps | ret(last 50)=+41.00 win_sr=100% cum_sr=100%]
... [9 sheep | 12,958,632 steps | ret(last 50)=+43.02 win_sr=100% cum_sr=100%]
... [9 sheep | 13,058,632 steps | ret(last 50)=+41.13 win_sr=100% cum_sr=100%]
... [9 sheep | 13,158,632 steps | ret(last 50)=+41.02 win_sr=100% cum_sr=100%]
... [9 sheep | 13,258,632 steps | ret(last 50)=+42.88 win_sr=100% cum_sr=100%]
... [9 sheep | 13,358,632 steps | ret(last 50)=+46.16 win_sr=100% cum_sr=100%]
... [9 sheep | 13,458,632 steps | ret(last 50)=+44.69 win_sr=100% cum_sr=100%]
... [9 sheep | 13,558,632 steps | ret(last 50)=+44.49 win_sr=100% cum_sr=100%]
[Stage n_sheep=9] evaluating 30 eps
[Stage n_sheep=9] sr=0% mean_len=4000 mean_min_pen=1.5m mean_act=1.39
failure modes: PARTIAL_8of9=26 PARTIAL_7of9=4
reward/step: progress=+0.0787 alignment=+0.0079 south=-0.0184 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0197 step_cost=-0.0200 complete=+0.0000
[Stage n_sheep=10] training 1,500,000 steps
... [10 sheep | 13,565,960 steps | ret(last 0)=+nan win_sr=nan% cum_sr=nan%]
... [10 sheep | 13,665,960 steps | ret(last 50)=+43.38 win_sr=100% cum_sr=100%]
... [10 sheep | 13,765,960 steps | ret(last 50)=+43.26 win_sr=100% cum_sr=100%]
... [10 sheep | 13,865,960 steps | ret(last 50)=+46.91 win_sr=100% cum_sr=100%]
... [10 sheep | 13,965,960 steps | ret(last 50)=+45.36 win_sr=100% cum_sr=100%]
... [10 sheep | 14,065,960 steps | ret(last 50)=+45.37 win_sr=100% cum_sr=100%]
... [10 sheep | 14,165,960 steps | ret(last 50)=+44.30 win_sr=100% cum_sr=100%]
... [10 sheep | 14,265,960 steps | ret(last 50)=+43.83 win_sr=100% cum_sr=100%]
... [10 sheep | 14,365,960 steps | ret(last 50)=+47.09 win_sr=100% cum_sr=100%]
... [10 sheep | 14,465,960 steps | ret(last 50)=+41.32 win_sr=100% cum_sr=100%]
... [10 sheep | 14,565,960 steps | ret(last 50)=+45.30 win_sr=100% cum_sr=100%]
... [10 sheep | 14,665,960 steps | ret(last 50)=+45.36 win_sr=98% cum_sr=100%]
... [10 sheep | 14,765,960 steps | ret(last 50)=+41.83 win_sr=100% cum_sr=100%]
... [10 sheep | 14,865,960 steps | ret(last 50)=+44.40 win_sr=100% cum_sr=100%]
... [10 sheep | 14,965,960 steps | ret(last 50)=+45.89 win_sr=100% cum_sr=100%]
... [10 sheep | 15,065,960 steps | ret(last 50)=+42.49 win_sr=100% cum_sr=100%]
[Stage n_sheep=10] evaluating 30 eps
[Stage n_sheep=10] sr=83% mean_len=2243 mean_min_pen=1.5m mean_act=1.40
failure modes: SUCCESS=25 PARTIAL_8of10=3 PARTIAL_7of10=2
reward/step: progress=+0.1387 alignment=+0.0150 south=-0.0437 compact=+0.0000 wall_touch=+0.0000 pen_bonus=+0.0428 step_cost=-0.0200 complete=+0.0372
======================================================================
TRAINING SUMMARY
======================================================================
n_sheep=1 sr=100% len= 264 min_pen= 3.7m act=0.45
n_sheep=2 sr= 77% len= 1398 min_pen= 3.3m act=0.97
n_sheep=3 sr= 97% len= 1095 min_pen= 2.5m act=0.95
n_sheep=4 sr= 57% len= 2572 min_pen= 2.2m act=1.28
n_sheep=5 sr= 0% len= 4000 min_pen= 1.7m act=1.36
n_sheep=6 sr= 37% len= 3137 min_pen= 1.8m act=1.37
n_sheep=7 sr= 0% len= 4000 min_pen= 1.8m act=1.38
n_sheep=8 sr= 60% len= 2472 min_pen= 1.6m act=1.39
n_sheep=9 sr= 0% len= 4000 min_pen= 1.5m act=1.39
n_sheep=10 sr= 83% len= 2243 min_pen= 1.5m act=1.40
Total time: 94.3 min
Artefacts: runs/v3/
Plots: runs/v3/success_rate.png, runs/v3/eval/