tonysy commented on issue #5199: add parallel actor critic algorithm
URL: https://github.com/apache/incubator-mxnet/pull/5199#issuecomment-332090447
 
 
   Hi, I have run this algorithm on my server(28  Intel(R) Xeon(R) CPU E5-2690 
v4 @ 2.60GHz), with the default parameter setting(num-envs = 16). It takes too 
much time to solve this problem, more than 10 hours.  I use the latest version 
of mxnet. I would like to know what cause this algorithm to be solved so 
slowly. Thansk.
   ```
   Batch 1061 complete (30.74s) (31801.8s elapsed) (episode 16992), batch avg. 
reward: 20.31, running reward:[938/1835]
   Batch 1062 complete (32.07s) (31833.8s elapsed) (episode 17008), batch avg. 
reward: 17.00, running reward: 16.872
   Batch 1063 complete (29.40s) (31863.2s elapsed) (episode 17024), batch avg. 
reward: 18.19, running reward: 17.056
   Batch 1064 complete (31.82s) (31895.0s elapsed) (episode 17040), batch avg. 
reward: 14.12, running reward: 16.611
   Batch 1065 complete (28.39s) (31923.4s elapsed) (episode 17056), batch avg. 
reward: 18.31, running reward: 16.843
   Batch 1066 complete (28.56s) (31952.0s elapsed) (episode 17072), batch avg. 
reward: 18.31, running reward: 17.088
   Batch 1067 complete (32.69s) (31984.7s elapsed) (episode 17088), batch avg. 
reward: 18.50, running reward: 17.277
   Batch 1068 complete (33.23s) (32017.9s elapsed) (episode 17104), batch avg. 
reward: 19.44, running reward: 17.599
   Batch 1069 complete (29.90s) (32047.8s elapsed) (episode 17120), batch avg. 
reward: 18.62, running reward: 17.774
   Batch 1070 complete (28.10s) (32075.9s elapsed) (episode 17136), batch avg. 
reward: 15.81, running reward: 17.524
   Batch 1071 complete (28.70s) (32104.6s elapsed) (episode 17152), batch avg. 
reward: 16.06, running reward: 17.347
   Batch 1072 complete (29.84s) (32134.5s elapsed) (episode 17168), batch avg. 
reward: 18.75, running reward: 17.563
   Batch 1073 complete (28.55s) (32163.0s elapsed) (episode 17184), batch avg. 
reward: 18.25, running reward: 17.668
   Batch 1074 complete (30.55s) (32193.6s elapsed) (episode 17200), batch avg. 
reward: 14.31, running reward: 17.162
   Batch 1075 complete (29.53s) (32223.1s elapsed) (episode 17216), batch avg. 
reward: 20.94, running reward: 17.723
   Batch 1076 complete (28.64s) (32251.7s elapsed) (episode 17232), batch avg. 
reward: 18.38, running reward: 17.826
   Batch 1077 complete (28.24s) (32280.0s elapsed) (episode 17248), batch avg. 
reward: 15.69, running reward: 17.507
   Batch 1078 complete (32.42s) (32312.4s elapsed) (episode 17264), batch avg. 
reward: 14.56, running reward: 17.091
   Batch 1079 complete (28.42s) (32340.8s elapsed) (episode 17280), batch avg. 
reward: 18.38, running reward: 17.302
   Batch 1080 complete (32.11s) (32372.9s elapsed) (episode 17296), batch avg. 
reward: 15.81, running reward: 17.074
   Batch 1081 complete (29.21s) (32402.1s elapsed) (episode 17312), batch avg. 
reward: 17.94, running reward: 17.221
   Batch 1082 complete (31.46s) (32433.6s elapsed) (episode 17328), batch avg. 
reward: 16.50, running reward: 17.131
   Batch 1083 complete (29.41s) (32463.0s elapsed) (episode 17344), batch avg. 
reward: 20.75, running reward: 17.670
   Batch 1084 complete (28.96s) (32492.0s elapsed) (episode 17360), batch avg. 
reward: 18.56, running reward: 17.790
   Batch 1085 complete (29.06s) (32521.0s elapsed) (episode 17376), batch avg. 
reward: 18.62, running reward: 17.933
   Batch 1086 complete (31.93s) (32553.0s elapsed) (episode 17392), batch avg. 
reward: 15.31, running reward: 17.507
   Batch 1087 complete (31.80s) (32584.7s elapsed) (episode 17408), batch avg. 
reward: 20.12, running reward: 17.890
   Batch 1088 complete (28.33s) (32613.1s elapsed) (episode 17424), batch avg. 
reward: 13.44, running reward: 17.215
   Batch 1089 complete (31.49s) (32644.6s elapsed) (episode 17440), batch avg. 
reward: 5.94, running reward: 15.559
   Batch 1090 complete (28.32s) (32672.9s elapsed) (episode 17456), batch avg. 
reward: 15.88, running reward: 15.631
   Batch 1091 complete (29.21s) (32702.1s elapsed) (episode 17472), batch avg. 
reward: 18.56, running reward: 16.079
   Batch 1092 complete (28.98s) (32731.1s elapsed) (episode 17488), batch avg. 
reward: 18.44, running reward: 16.443
   Batch 1093 complete (29.88s) (32761.0s elapsed) (episode 17504), batch avg. 
reward: 18.06, running reward: 16.655
   Batch 1094 complete (29.35s) (32790.3s elapsed) (episode 17520), batch avg. 
reward: 18.12, running reward: 16.868
   Batch 1095 complete (29.36s) (32819.7s elapsed) (episode 17536), batch avg. 
reward: 12.88, running reward: 16.280
   Batch 1096 complete (29.65s) (32849.3s elapsed) (episode 17552), batch avg. 
reward: 18.75, running reward: 16.636
   Batch 1097 complete (31.78s) (32881.1s elapsed) (episode 17568), batch avg. 
reward: 19.62, running reward: 17.077
   Batch 1098 complete (29.47s) (32910.6s elapsed) (episode 17584), batch avg. 
reward: 15.12, running reward: 16.774
   Batch 1099 complete (30.07s) (32940.6s elapsed) (episode 17600), batch avg. 
reward: 20.88, running reward: 17.382
   Batch 1100 complete (30.22s) (32970.9s elapsed) (episode 17616), batch avg. 
reward: 15.62, running reward: 17.124
   Batch 1101 complete (33.70s) (33004.6s elapsed) (episode 17632), batch avg. 
reward: 17.31, running reward: 17.167
   Batch 1102 complete (32.87s) (33037.4s elapsed) (episode 17648), batch avg. 
reward: 14.56, running reward: 16.810
   Batch 1103 complete (29.89s) (33067.3s elapsed) (episode 17664), batch avg. 
reward: 18.62, running reward: 17.073
   Batch 1104 complete (29.67s) (33097.0s elapsed) (episode 17680), batch avg. 
reward: 18.44, running reward: 17.298
   Batch 1105 complete (30.19s) (33127.2s elapsed) (episode 17696), batch avg. 
reward: 16.06, running reward: 17.145
   Batch 1106 complete (30.90s) (33158.1s elapsed) (episode 17712), batch avg. 
reward: 14.06, running reward: 16.678
   Batch 1107 complete (34.82s) (33192.9s elapsed) (episode 17728), batch avg. 
reward: 17.19, running reward: 16.770
   Batch 1108 complete (29.23s) (33222.1s elapsed) (episode 17744), batch avg. 
reward: 17.94, running reward: 16.962
   Batch 1109 complete (28.70s) (33250.8s elapsed) (episode 17760), batch avg. 
reward: 18.25, running reward: 17.158
   Batch 1110 complete (28.27s) (33279.1s elapsed) (episode 17776), batch avg. 
reward: 15.50, running reward: 16.960
   Batch 1111 complete (29.08s) (33308.2s elapsed) (episode 17792), batch avg. 
reward: 13.62, running reward: 16.446
   Batch 1112 complete (32.80s) (33341.0s elapsed) (episode 17808), batch avg. 
reward: 14.88, running reward: 16.227
   Batch 1113 complete (32.26s) (33373.2s elapsed) (episode 17824), batch avg. 
reward: 17.19, running reward: 16.361
   Batch 1114 complete (33.19s) (33406.4s elapsed) (episode 17840), batch avg. 
reward: 19.94, running reward: 16.898
   Batch 1115 complete (31.45s) (33437.9s elapsed) (episode 17856), batch avg. 
reward: 16.62, running reward: 16.832
   Batch 1116 complete (32.06s) (33469.9s elapsed) (episode 17872), batch avg. 
reward: 13.75, running reward: 16.413
   Batch 1117 complete (34.21s) (33504.1s elapsed) (episode 17888), batch avg. 
reward: 17.62, running reward: 16.565
   Batch 1118 complete (31.35s) (33535.5s elapsed) (episode 17904), batch avg. 
reward: 16.19, running reward: 16.543
   Batch 1119 complete (29.52s) (33565.0s elapsed) (episode 17920), batch avg. 
reward: 20.75, running reward: 17.167
   Batch 1120 complete (33.63s) (33598.7s elapsed) (episode 17936), batch avg. 
reward: 13.94, running reward: 16.722
   Batch 1121 complete (30.43s) (33629.1s elapsed) (episode 17952), batch avg. 
reward: 15.81, running reward: 16.578
   Batch 1122 complete (28.45s) (33657.5s elapsed) (episode 17968), batch avg. 
reward: 7.81, running reward: 15.319
   Batch 1123 complete (29.82s) (33687.4s elapsed) (episode 17984), batch avg. 
reward: 18.00, running reward: 15.710
   Batch 1124 complete (33.27s) (33720.6s elapsed) (episode 18000), batch avg. 
reward: 18.81, running reward: 16.157
   Batch 1125 complete (28.94s) (33749.6s elapsed) (episode 18016), batch avg. 
reward: 15.50, running reward: 16.049
   Batch 1126 complete (32.01s) (33781.6s elapsed) (episode 18032), batch avg. 
reward: 17.56, running reward: 16.271
   Batch 1127 complete (33.79s) (33815.4s elapsed) (episode 18048), batch avg. 
reward: 16.75, running reward: 16.298
   Batch 1128 complete (29.23s) (33844.6s elapsed) (episode 18064), batch avg. 
reward: 20.81, running reward: 16.969
   Batch 1129 complete (34.03s) (33878.6s elapsed) (episode 18080), batch avg. 
reward: 12.38, running reward: 16.258
   Batch 1130 complete (29.88s) (33908.5s elapsed) (episode 18096), batch avg. 
reward: 20.50, running reward: 16.888
   Batch 1131 complete (27.83s) (33936.3s elapsed) (episode 18112), batch avg. 
reward: 10.19, running reward: 15.899
   Batch 1132 complete (29.85s) (33966.2s elapsed) (episode 18128), batch avg. 
reward: 19.12, running reward: 16.382
   Batch 1133 complete (28.03s) (33994.2s elapsed) (episode 18144), batch avg. 
reward: 15.69, running reward: 16.316
   Batch 1134 complete (28.33s) (34022.5s elapsed) (episode 18160), batch avg. 
reward: 13.44, running reward: 15.900
   Batch 1135 complete (32.37s) (34054.9s elapsed) (episode 18176), batch avg. 
reward: 19.00, running reward: 16.374
   Batch 1136 complete (30.28s) (34085.2s elapsed) (episode 18192), batch avg. 
reward: 16.50, running reward: 16.376
   Batch 1137 complete (29.31s) (34114.5s elapsed) (episode 18208), batch avg. 
reward: 20.81, running reward: 17.036
   Batch 1138 complete (28.40s) (34142.9s elapsed) (episode 18224), batch avg. 
reward: 12.81, running reward: 16.432
   Batch 1139 complete (29.99s) (34172.9s elapsed) (episode 18240), batch avg. 
reward: 13.88, running reward: 16.019
   Batch 1140 complete (29.17s) (34202.0s elapsed) (episode 18256), batch avg. 
reward: 18.12, running reward: 16.307
   Batch 1141 complete (28.81s) (34230.9s elapsed) (episode 18272), batch avg. 
reward: 18.06, running reward: 16.593
   Batch 1142 complete (28.09s) (34258.9s elapsed) (episode 18288), batch avg. 
reward: 15.75, running reward: 16.500
   ........
   .........
   Batch 1722 complete (28.69s) (51759.5s elapsed) (episode 27568), batch avg. 
reward: 21.00, running reward: 18.450
   Batch 1723 complete (30.06s) (51789.5s elapsed) (episode 27584), batch avg. 
reward: 17.88, running reward: 18.347
   Batch 1724 complete (30.61s) (51820.1s elapsed) (episode 27600), batch avg. 
reward: 16.62, running reward: 18.099
   Batch 1725 complete (29.28s) (51849.4s elapsed) (episode 27616), batch avg. 
reward: 20.88, running reward: 18.512
   Batch 1726 complete (29.61s) (51879.0s elapsed) (episode 27632), batch avg. 
reward: 20.75, running reward: 18.845
   Batch 1727 complete (28.95s) (51908.0s elapsed) (episode 27648), batch avg. 
reward: 18.44, running reward: 18.783
   Batch 1728 complete (29.00s) (51937.0s elapsed) (episode 27664), batch avg. 
reward: 20.88, running reward: 19.095
   Batch 1729 complete (29.87s) (51966.8s elapsed) (episode 27680), batch avg. 
reward: 20.62, running reward: 19.324
   Batch 1730 complete (29.19s) (51996.0s elapsed) (episode 27696), batch avg. 
reward: 20.94, running reward: 19.564
   Batch 1731 complete (30.47s) (52026.5s elapsed) (episode 27712), batch avg. 
reward: 20.62, running reward: 19.722
   ```
 
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to