H H q;aqA
brain.Event:2R.
,tensorboard.summary.writer.event_file_writer) L" x= utqA*
objective/kl ks' F DvtqA*
objective/kl_coefL>#' F ivtqA*
objective/entropy6B#J/ m] P vtqA*"
ppo/mean_non_score_reward oV% 6 vtqA*
ppo/mean_scores=}M$ B+M vtqA*
ppo/std_scores=- |'( pJ ?}tqA*
ppo/policy/entropy-V@) 7_ R}tqA*
ppo/policy/approxkl?FO*) 7_ ytqA*
ppo/policy/policyklww>|T) 7_ tqA*
ppo/policy/clipfrac>0 _ tqA*#
!
ppo/policy/advantages_mean43w3 $& sO ՂtqA*
ppo/returns/meant?Z% 6 tqA*
ppo/returns/var1>,# wC tqA*
ppo/val/vpredc$C@e=}# wC tqA*
ppo/val/error M@V& sO tqA*
ppo/val/clipfrac ?q~1" x= tqA*
ppo/val/mean@Z@Pq ! { tqA*
ppo/val/var$_(@+ K ftqA*
ppo/val/var_explainedt74' F @tqA*
ppo/learning_rate
l7p9+ K tqA*
time/ppo/forward_pass0B>cЌ. W tqA*!
time/ppo/compute_rewards x;
=1 RtqA*$
"
time/ppo/compute_advantages P;, E tqA*
time/ppo/optimize_stepʁ?-) 7_ tqA*
time/ppo/calc_statspb?>r$ B+M 0tqA*
time/ppo/totalݳ?.% 6 ύtqA*
env/reward_mean=;+3$ B+M mtqA*
env/reward_std=d" x= WqA*
objective/kl?@
' F
XqA*
objective/kl_coef?L>E' F )XqA*
objective/entropy*`B\b9/ m] P ?XqA*"
ppo/mean_non_score_rewardƼsTD% 6 SXqA*
ppo/mean_scores=,$ B+M fXqA*
ppo/std_scores
V=S- V?( pJ aqA*
ppo/policy/entropyQ@|5) 7_ (aqA*
ppo/policy/approxkl> }&# wC JgqA*
ppo/val/vpred"@_%# wC iqA*
ppo/val/error@& sO iqA*
ppo/val/clipfrac43>0pw" x= kqA*
ppo/val/meanT3@H! { lqA*
ppo/val/var?г+ K 1lqA*
ppo/val/var_explained/*gݟD' F nqA*
ppo/learning_rate
l75!L{+ K FnqA*
time/ppo/forward_passdB>omLJ. W _nqA*!
time/ppo/compute_rewards :!1 nqA*$
"
time/ppo/compute_advantages L$ B+M pqA*
time/ppo/total֧?#,% 6 rqqA*
env/reward_mean=g$ B+M rqA*
env/reward_std
V=Z`" x= tqA*
objective/kl@̃' F RtqA*
objective/kl_coefL>sqM' F utqA*
objective/entropyfB/ m] P tqA*"
ppo/mean_non_score_reward@`% 6 tqA*
ppo/mean_scores =Zȣ$ B+M tqA*
ppo/std_scoresm+= 08- db^V( pJ htqA*
ppo/policy/entropyJ@w) 7_ tqA*
ppo/policy/approxklI; ) 7_ +tqA*
ppo/policy/policykl@R;$i) 7_ tqA*
ppo/policy/clipfracgY=B0 _ tqA*#
!
ppo/policy/advantages_mean.K3$ߎ & sO _ # wC tqA*
ppo/val/vpredT`?2p# wC tqA*
ppo/val/errorI`@<& sO tqA*
ppo/val/clipfrac >vK(" x= tqA*
ppo/val/mean@&i! { tqA*
ppo/val/var&?ҹG+ K tqA*
ppo/val/var_explainedrI"' F btqA*
ppo/learning_rate
l7\+ K StqA*
time/ppo/forward_passB>m. W tqA*!
time/ppo/compute_rewards @: b1 tqA*$
"
time/ppo/compute_advantages A;m, E tqA*
time/ppo/optimize_steplo?I) 7_ tqA*
time/ppo/calc_stats 2>s$ B+M tqA*
time/ppo/total(?8% 6
tqA*
env/reward_mean = $ B+M tqA*
env/reward_stdm+=v-k" x= 7qA*
objective/klW@гѤ' F a7qA*
objective/kl_coef3hL>M4' F 7qA*
objective/entropy\B"/ m] P 7qA*"
ppo/mean_non_score_reward % 6 7qA*
ppo/mean_scoresT=$ B+M 7qA*
ppo/std_scores =";- + K '7qA*
time/ppo/forward_passbD>&. W '7qA*!
time/ppo/compute_rewards P:3mؑ1 '7qA*$
"
time/ppo/compute_advantages 0@;, E '7qA*
time/ppo/optimize_step"?eT) 7_ '7qA*
time/ppo/calc_statsH1>y$ B+M '7qA*
time/ppo/totalݮ?pKt% 6 ^(7qA*
env/reward_meanT=_ņ$ B+M (7qA*
env/reward_std =#]" x= \qA*
objective/klG @' F }\qA*
objective/kl_coefFL>Q:' F \qA*
objective/entropyXYBq/ m] P \qA*"
ppo/mean_non_score_reward% 6 \qA*
ppo/mean_scoresއ=P$?k$ B+M \qA*
ppo/std_scoresf=WT- (r% 6 dqA*
ppo/returns/varU*># wC dqA*
ppo/val/vpredC?q
# wC XeqA*
ppo/val/error=@yF& sO eqA*
ppo/val/clipfracg>[ " x= fqA*
ppo/val/meanR?A k! { dfqA*
ppo/val/varƷ?jK+ K fqA*
ppo/val/var_explained Tpˤ7' F gqA*
ppo/learning_rate
l7q
D+ K wgqA*
time/ppo/forward_passzC> . W gqA*!
time/ppo/compute_rewards :L1 3hqA*$
"
time/ppo/compute_advantages F;>S, E hqA*
time/ppo/optimize_stepL~?Z ) 7_ hqA*
time/ppo/calc_statsP1> h$$ B+M =iqA*
time/ppo/total?q% 6 iqA*
env/reward_meanއ=鑟n$ B+M iqA*
env/reward_stdf=Z" x= .qA*
objective/kl
m@?' F R.qA*
objective/kl_coef>%L>z ' F o.qA*
objective/entropyܾB/ m] P .qA*"
ppo/mean_non_score_reward4&+% 6 .qA*
ppo/mean_scores:=tS$ B+M .qA*
ppo/std_scores s=!- {% 6 .qA*
ppo/returns/var@>2\ # wC .qA*
ppo/val/vpred)8?Z\# wC .qA*
ppo/val/error*?{
& sO .qA*
ppo/val/clipfracffD>X\" x= .qA*
ppo/val/meanμU?e/! { .qA*
ppo/val/var)?+ K ).qA*
ppo/val/var_explained[8' F 9.qA*
ppo/learning_rate
l7SL + K I.qA*
time/ppo/forward_passPIC>. W Z.qA*!
time/ppo/compute_rewards `:[<91 j.qA*$
"
time/ppo/compute_advantages E;'}, E .qA*
time/ppo/optimize_step~?ybϫ) 7_ .qA*
time/ppo/calc_statsPp2>l$ B+M ).qA*
time/ppo/totalҮ?% 6 .qA*
env/reward_mean:=4]$ B+M a.qA*
env/reward_std s=*" x= 0wqA*
objective/klE,@q' F wqA*
objective/kl_coefL>Z' F wqA*
objective/entropyjB0uo/ m] P wqA*"
ppo/mean_non_score_rewardۼ=Y% 6 wqA*
ppo/mean_scoresw=dq$ B+M wqA*
ppo/std_scoresk<>- ,C# wC EwqA*
ppo/val/vpredy>SJC# wC wqA*
ppo/val/errorJ?We& sO zwqA*
ppo/val/clipfrac33 >S" x= wqA*
ppo/val/meanjx? T! { wqA*
ppo/val/varʟ?R*+ K NwqA*
ppo/val/var_explained5(' F wqA*
ppo/learning_rate
l7n+ K wqA*
time/ppo/forward_passC>.I9. W 5wqA*!
time/ppo/compute_rewards :D|1 wqA*$
"
time/ppo/compute_advantages >;yg, E fwqA*
time/ppo/optimize_step$Q?) 7_ wqA*
time/ppo/calc_stats 2>_\$ B+M wqA*
time/ppo/total6?4% 6 ,wqA*
env/reward_meanw=yT$ B+M wqA*
env/reward_stdk<>ﻛ|" x= LqA*
objective/klì?4E[' F \MqA*
objective/kl_coef_K>Sk' F MqA*
objective/entropy)BJu/ m] P MqA*"
ppo/mean_non_score_reward&\w% 6 MqA*
ppo/mean_scores =ގ&$ B+M MqA*
ppo/std_scoresh=tOJ- =Q6% 6 G^qA*
ppo/returns/var$># wC q^qA*
ppo/val/vpredqY>h# wC uaqA*
ppo/val/error ?ů*Z& sO ٘D! { fqA*
ppo/val/var?+ K gqA*
ppo/val/var_explainedtaמ' F 8iqA*
ppo/learning_rate
l7ֽ+ K eiqA*
time/ppo/forward_passB>. W |iqA*!
time/ppo/compute_rewards :Ro1 ckqA*$
"
time/ppo/compute_advantages E;MB, E kqA*
time/ppo/optimize_step~?{) 7_ kqA*
time/ppo/calc_stats'2>8I$ B+M UlqA*
time/ppo/total?FGa% 6 lqA*
env/reward_mean =Ѧb$ B+M mqA*
env/reward_stdh=1" x=
qA *
objective/kl?1Y?' F b
qA *
objective/kl_coefK>' F
qA *
objective/entropy B|/ m] P
qA *"
ppo/mean_non_score_reward{ƙ:u% 6
qA *
ppo/mean_scores=ȭ$ B+M
qA *
ppo/std_scores-p=K- $ B+M R
qA *
ppo/loss/totalAR<f( pJ
qA *
ppo/policy/entropyzw@t ) 7_
qA *
ppo/policy/approxklʬ;0) 7_
qA *
ppo/policy/policykl
Z;
l) 7_ ݚ
qA *
ppo/policy/clipfracY=fu+0 _
qA *#
!
ppo/policy/advantages_meanffvy& sO
qA *
ppo/returns/mean2=~{{% 6 .
qA *
ppo/returns/varW=D# wC G
qA *
ppo/val/vprede><# wC Z
qA *
ppo/val/errorp?jM& sO
qA *
ppo/val/clipfrac;/y'" x=
qA *
ppo/val/meanK> ! {
qA *
ppo/val/var:j?Pi+ K i
qA *
ppo/val/var_explainedI5@' F
qA *
ppo/learning_rate
l7@+ K
qA *
time/ppo/forward_pass0 B>#XS. W [
qA *!
time/ppo/compute_rewards :&E W1
qA *$
"
time/ppo/compute_advantages I;$<, E
qA *
time/ppo/optimize_step0~?,,k) 7_ <
qA *
time/ppo/calc_stats e2>#F/$ B+M ڨ
qA *
time/ppo/totalF?ؠ.% 6 z
qA *
env/reward_mean=A/$ B+M
qA *
env/reward_std-p=r" x= bWrA
*
objective/kla]?T @+' F wbWrA
*
objective/kl_coefK>|' F bWrA
*
objective/entropy쓝Bk / m] P cWrA
*"
ppo/mean_non_score_rewardP~<% 6 cWrA
*
ppo/mean_scores;=
-$ B+M 1cWrA
*
ppo/std_scoresR=S,- # wC rWrA
*
ppo/val/vpred'>[# wC rWrA
*
ppo/val/error?"[ & sO uWrA
*
ppo/val/clipfrač;?&" x= uWrA
*
ppo/val/meanH>]! { 'yWrA
*
ppo/val/var?
!+ K RyWrA
*
ppo/val/var_explainedƹ~ ' F iyWrA
*
ppo/learning_rate
l7F+ K {yWrA
*
time/ppo/forward_pass A>j.. W zWrA
*!
time/ppo/compute_rewards :'-1 {WrA
*$
"
time/ppo/compute_advantages E;@c, E s|WrA
*
time/ppo/optimize_stepP~?9$N) 7_ }WrA
*
time/ppo/calc_stats1>]y_o$ B+M }WrA
*
time/ppo/total?
:% 6 i~WrA
*
env/reward_mean;=f$ B+M WrA
*
env/reward_stdR= " x= rA *
objective/klq.d@ v/' F ] rA *
objective/kl_coef8~K>
J' F ~ rA *
objective/entropyBQ
q/ m] P rA *"
ppo/mean_non_score_reward´% 6 rA *
ppo/mean_scores=xpC$ B+M rA *
ppo/std_scores4=1- j$ B+M rA *
ppo/loss/value ?>e$ B+M rA *
ppo/loss/total# wC _ rA *
ppo/val/vpred-> Q# wC rA *
ppo/val/error4I?& sO X rA *
ppo/val/clipfracff;ܜ" x= rA *
ppo/val/meanI>c3! { rA *
ppo/val/varB!?pq + K g rA *
ppo/val/var_explainedcV_?' F 5 rA *
ppo/learning_rate
l7{>X+ K ۜ rA *
time/ppo/forward_passpB> . W rA *!
time/ppo/compute_rewards :lۖ1 & rA *$
"
time/ppo/compute_advantages ;\1, E Ǟ rA *
time/ppo/optimize_stepK? ) 7_ d rA *
time/ppo/calc_stats3>̚W$ B+M rA *
time/ppo/total*?Y!% 6 rA *
env/reward_mean=i5$ B+M < rA *
env/reward_std4=H(o" x= tP/rA *
objective/kl?[$' F P/rA *
objective/kl_coef\K>[' F iP/rA *
objective/entropy aBY/ m] P P/rA *"
ppo/mean_non_score_rewardeai% 6 P/rA *
ppo/mean_scores =nH$ B+M P/rA *
ppo/std_scoresf=}V - f$ B+M P/rA *
ppo/loss/value7>ܜW$ B+M xP/rA *
ppo/loss/totalD PG# wC WP/rA *
ppo/val/vpredj>. W Q/rA *!
time/ppo/compute_rewards :Hv1 IQ/rA *$
"
time/ppo/compute_advantages HJ;psK, E Q/rA *
time/ppo/optimize_stepr?k) 7_ Q/rA *
time/ppo/calc_stats@N1>`b$ B+M ,Q/rA *
time/ppo/totalZ?ǝ@% 6 Q/rA *
env/reward_mean =>F$ B+M jQ/rA *
env/reward_stdf=tW