BC03 3DoF, 1 Sphere, PN++ averaging layer (debugging, prior data)

Using older data (ladle_algorithmic_v02) to debug

Created on May 16|Last edited on May 20

Comment

Well this is working well. 
TL;DR:
This is with the v02 3DoF data.
"bc_data_dir": "/data/seita/softgym_mm/data_demo/MMOneSphere_v01_BClone_filtered_ladle_algorithmic_v02_nVars_2000_obs_combo_act_translation",
I did a batch of 4X runs on 05/14/2022 (might have been 5 except 1 could have crashed) for 250 epochs, then 1X run on 05/20/2022 for 300 epochs.
These are all after the bug fix on 05/14 which fixed evaluation and thus we should be able to trust the evaluation MSEs, etc.
Can compare with "similar" translation-only v02 3DoF data here: https://wandb.ai/mooey5775/mixed_media/reports/BC03-3DoF-1-Sphere-PN-Average-Layer-EE--VmlldzoyMDAzMjcz﻿
Section 1﻿
PN++ Avg (v02 data)5
﻿
GIFsSeed 01 after 250. LGTM?
﻿
Seed 02 after 250. It must be jittering around in water and I thought the newer demonstrator would help with that, as ladle_algorithmic_v08 is designed to move up AS SOON AS the condition is satisfied, whereas the current demonstrator can go back and forth.
﻿
Seed 03 after 250, LGTM.
﻿
OK this one had 0/10 but that could be due to the demonstrator or due to the ball getting sunk (so it is not seen in the point cloud).
﻿
﻿

Add a comment