Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

train MiniRTS is slow - no information when will end #128

Open
xothor opened this issue Apr 20, 2019 · 1 comment
Open

train MiniRTS is slow - no information when will end #128

xothor opened this issue Apr 20, 2019 · 1 comment

Comments

@xothor
Copy link

xothor commented Apr 20, 2019

I started training miniRTS - like 2 weeks ago. Without interrupting, for now I have (just the last part pasted):

[trainer] actor count: 4750/5000
new_record: False
count: 2905
best_win_rate: 0.7521728917077578
str_win_rate: [2905] Win rate: 0.727 [3103/1165/4268], Best win rate: 0.752 [2754]
str_acc_win_rate: Accumulated win rate: 0.667 [8208765/4101445/12310210]
100%|█████████| 5000/5000 [07:34<00:00, 11.00it/s]
[2019-04-20 14:31:03.794200][64] Iter[2906]: 
Save to ./
Filename = ./save-726740.bin
Command arguments train.py --batchsize 64 --freq_update 1 --players type=AI_NN,fs=50,args=backup/AI_SIMPLE|start/500|decay/0.99;type=AI_SIMPLE,fs=20 --num_games 1024 --tqdm --T 20 --additional_labels id,last_terminal --trainer_stats winrate --keys_in_reply V --gpu 0
[2906] Time spent = 454386.121000 ms
actor: 4750/5000
train: 250/5000
2906:V_err[4750]: avg: 0.03603, min: 0.00120[3002], max: 0.13680[1537]
2906:acc_reward[4750]: avg: 0.33279, min: 0.12162[3137], max: 0.51011[116]
2906:cost[250]: avg: 1.84572, min: 1.75103[225], max: 1.91308[153]
2906:entropy_pi[4750]: avg: -1.81812, min: -1.97498[4702], max: -1.58594[3505]
2906:init_reward[250]: avg: 0.33657, min: 0.18084[165], max: 0.47968[192]
2906:nll_pi[4750]: avg: 1.82787, min: 1.50373[3047], max: 2.29496[1670]
2906:predicted_V[4750]: avg: 0.33951, min: -1.25702[2216], max: 1.27786[2678]
2906:reward[4750]: avg: 0.00634, min: -0.04688[1216], max: 0.06250[1082]
2906:total_entropy[4750]: avg: -1.81812, min: -1.97498[4702], max: -1.58594[3505]
2906:total_nll[4750]: avg: 1.82787, min: 1.50373[3047], max: 2.29496[1670]

[trainer] actor count: 4750/5000
new_record: False
count: 2906
best_win_rate: 0.7521728917077578
str_win_rate: [2906] Win rate: 0.724 [3070/1173/4243], Best win rate: 0.752 [2754]
str_acc_win_rate: Accumulated win rate: 0.667 [8211835/4102618/12314453]

For now I do not now if there is any 'soft option to finish' in current state to get result for research.

Info about workstation:
Memory: 8 GiB
CPU: Intel Core i5-4590 3.30GHz x4
FeForce GTX 960 2GB GDDR5 (128 bit)
OS: Ubuntu 18.04.2 LTS

@lafmdp
Copy link

lafmdp commented May 3, 2019

Maybe you can control it at ./rlpytorch/runner/single_process.py

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants