From 9b54b72a2537a3225700e11c74a84e5bed1b2b6e Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jan=20L=C3=B6wenstrom?= Date: Tue, 3 Mar 2020 02:52:39 +0100 Subject: [PATCH] add epsilon convergence test and will remove unnecessary multithreaded learning --- Rplot.png | Bin 0 -> 7659 bytes convergence.txt | 20 +++++ src/main/java/core/algo/EpisodicLearning.java | 78 +++++++++++------- src/main/java/core/algo/Learning.java | 6 +- .../MonteCarloControlFirstVisitEGreedy.java | 36 ++++++-- .../java/core/controller/RLController.java | 28 ++++--- src/main/java/example/JumpingDino.java | 48 ++++++++--- src/main/java/example/Results | 15 ++++ 8 files changed, 167 insertions(+), 64 deletions(-) create mode 100644 Rplot.png create mode 100644 convergence.txt create mode 100644 src/main/java/example/Results diff --git a/Rplot.png b/Rplot.png new file mode 100644 index 0000000000000000000000000000000000000000..dc1e2e49536ed7d5a1f8ebcfca38b02fedbfc870 GIT binary patch literal 7659 zcmeHMc~Fx{l#j~j-~poyprD{~L^(tS#6%893lG#&)U z0}ntC3W7s{UpyVYH3tWN@$qmx9**b7LqYLyP&__9p3cX^L3nyRo{z`#`Atnt^mr(p z&*y_cd=Qimg7cvuJ`~D_0-bn19K?si`CIgSfMSbSKnGCLGaSR}r z@O>j&BnTwZD)?_}M-<%wfevpNU-`otjr%dUFLa=1$y& z&igyre7-}%hp2Z(cIEMZT;ha~KKSR@Y!{YMQ&mk-R0jR2*c%E0MaV(6<;a6CCNqT| zC*nYsiDE}he79dQHh*w}DEynk|B*>?Q*vs%@hq<#!M+E2bTk=PNo*g*uB}Kv*dHy= z>KHqv<|^Cd$G-)7bSx3q5oJL~u*kjP3*8e!k9V&iy`?hNA>n6Y9PMGczxVc3pASDX zO+u5~YIvz_R4O%&#y!byN6AXO^9V5}k5hr+wK;loW>~|3-1egGq=b2j1%DIIUqsTo z9CUfKg-0NAFEAP=Q8?L_nmW8W#9d~`d28+HiX3`sqO^=W)f^hOo*v2S2rx5$85N{{ zpeY#_gYrDp{DUv@61Ltq(Wz2tz+%s7{sQXl#wN@cCbA2U8DErLhRb74LMpS3Vr1(f z;Z0K=XmZ3k7TZdnpeaT#l2pDE#!|%FtSO zFQ*y;Tx{Ac!I1F%C-eu`4|7{nCj(`TL={-_Q>9TmyPk(gk%mJuLsly==ar$iiEClE z<-uK|4H}f=aWu(EX70L8YQ@&jNcA$2ImjAlV^zObf`nVvPN2x??N$JE?8UgRSZ?*0*JaPy}Olg`?%A6*}Vq-Ll^2Z?pH zoI2ElpnMHGhE^PFZ|Nx)7l3J^rSVoQxn~)u#Z^}x)dz$S$p1-Shce#U$SL;W+&_I2xe>~G-}u5*qC^Vr-O=apbKWn4)NB7CtXy8du=kh0 zQ6Zv#hARp!n=U(Lp=&*0_fv3h#2JJK>I`tT3m^4=I94I%Lj06KOU2v z9J^Zpi{*`llKTV4MgW1fNA>oQghjhL0ft51-S;ZQiW7m&37d>{CN2{|>*>Y7vb?yp ztSs`Oy=s6l(VBX${vib3JNYd;Kt@<|VLG=@H!_6`zB$Wd6g`k8<`1O6>kvv^`6Iy~ zTAo{97fAjQA=u5JWycl56l>-mof(KDM~nR;zn4CNz+Ruay9H#!&K+0kE(vX=B8KL{ zqn{lkw&naMoxfqYQI8)ARa_M-F}1tpyFH@HB6o6)fCDs*jL=ATMJi;_mXdMremz|H zr6V_C9<*EqWI2&YSwYEq9n;2|Yf0K;~$1vO&y7fkm0Lu;?3T zJ#RZGP>{r7b-Vsu?d0R%O?q6dt3#HeYoq>mG!OW zVcF}bo2hREUPm-S3$!u=CpBO+Ftn2dz!EQ$4XC29Vl^JKqQsCoGx6eiQlP+FOpqQhLVBrsaX#wG=PW8 zw-CDl*o-c4s6MoLZ_l=;fHq10;gJ%pF+jfM5N!BuiI#s#1rq*;Ad&uk`nc06Yd*cS zS6N~&GXVs@hQh~S(YA^1@N>^fPq_(PqPv7yn$~~5;&0Hf8tJp`$ezI#RTS#U&oHWn zT(JA;+wwM#K{+dK$-@e{F4ot_&I*jx&FjIc?IYncUb2gau5FnibUHTmkx4+KU$3+C z2NR+#O0R?VM8DV|M1Y9D+tMPsx2hD72A*ZW{){b}>%_-=X94NY51)B@tld95T7rd2 zL@q2XEy?jG=*3IFe0$A!hbFJhVG*Gd33V7Ci0Jjzt zcnxhaO8LsW$VQXgbOYCAqgCEvRJK<@B5qA?W1T~3mIvQIR^D9%hmVEgA?h9F{N>~g z4+TiLRshbRBPu=az_PDA7jq zfighcG}m)wtc`bPv1}?+Ux48?sqyz~aTLv;K6BhH*+LrjGb{erwJGW#uM@LH~a zMc(p(y3c-utG?zr$K!UtB>v$G?2>U)q>1ZvKu44)%w10~{CO*n zEsEz58z{mY5p##XNB#(1zwW!;Mo`9>WLM}q^5^f`omqfmP{Lb|vXXr^j1!pRcnqox z738CaYea)u*Q~u~m4siue~%`sK&A~_z~%qA9dOF5_)8K0PAh&4N?n67y{^R6R*+@< z&|U;Mo-;pnU@KoAqWr9$qObK28S0D;Kra%y#dD0$?QHTTkA9!J{xhi?D2P|e-xvhT z)&A|wyW~dHfvl);{-AKyt0Thv#l9{{;H=tzf`f#|l5HLJSe}bM4W4a{M+H_GOPF6) zB|ADfamwB&iq#ey6?ryXvY`>O1(QZa3_C1xb7^ULZU);{!;PxeJ02%{u;hA?$}Cp5 zrU@&;R9uHhPWR!$I+`}tyx3Ju`ZBtw?b_u=rdcKfp*=+vT7E{vG`;N7FA)A~qt08+(;^nzq!_{sGb!Tr;T?0aJ;t%;bq;eWJ>0LHk7;D`zTZ@Q7CT` zF3%11r*z@VvAF(&dHSK2ClJV~38Cw;f%057)#^)i8kkjaP+YX@o~%+wR9uNIZ&xqE zLtgu$>BhCR(AA3CW_z^7ia&{SMcACxkw#N+|WOH8MTD62fk| zU5Hxj;DED|qgY;uq zINP1@3~UFYWQ{6Ie`~N0PsqkXl4DK&dLKlaXq;@=8SorOq31bh9p>DQP3JNgHMfz^ znlQjYx3d`7NCoOK_f0|&!veXI*nz3&>EAlZwS!M~e9YJAa2WHGNuVjH{?trzgR<=| z=gj(=;il?onImCW;ai9dG1^Z^3zrjxbB{x;ra+%X3|qoh&%UnJv=@Xr>;y;w-1Ryl zVZPGSe6=#?=NbNd>5OcGAC&`C^cN)sxksTeVIB++J5@tM))%h~9=gCRR{v`|Vyrfc zYP4#Fa-E5fLF`8dw^w#%Xx&I;YO^6G<_CqwfE@?2`~W5UUA1d!Lg`nHMN377rQh1h zQuYhpD;~wU4te@IHO_bWUVwb!+OHYhKOHpA`8Yt)Rb+Wq9GZb-SwCG*#yN`1phMRK z7J@!msR#Nh)&}mHi$!VhdRm#Cr8%wY{8N9?~bk}$e_P2M>u21ymJsxx(U!eY`zTi=8sRI<)l zv78<(;7+1+jLA95uKV53NoavLo&4^fT0$=NET&Esn1)c_hlcN)J+*DAlGOiWO5W9M zBq2u!lK zZ(GVdVMcUm?}BsC4g~1-oN=|ZQ(E)al_N7Gk>de+O@wk2^-!iw$V8F_ElC=nsB0;! zx=N1FdhnK$w~o{UF%9b?2wz;7vtd&u-73g{wO-Gryy!g_4HO*Tyis0@3W{m)^G5Wf zjegq~e`Y091N+CNeq&BqgwFdBiy3z-}x!GInGE0*ip_9~3LOgaiBsPPz z0wx(4;ooP=PyW(AG`Id^BeYvTa?xe)G7=$&;KkH`eV^%2`zr_O)c(HnV_GT6HUT$) zlT>M}@0NBJmkn-STV!))nrm)~On#y5E>5^4w?taQvQ7w&*_`cf+r7bfG+n*))c?SU|^2iU}V%{ITn{7rj#V z8?$kB%fz(jL!gDX9IRS`b`Z_POjY~W$#*nHYyDi?Qpl5(Y3dBinLW4s50iT_*-svR z=HU|BgPbVqx*iIyq?WhVhirVEcs+=Gt&5l$o>C^R-0sy|&Q!r^Uro;G8%>6~GZn9+ zpW0{-{dw=D4ZRTkrmgGBr6}Usiq4^^1bZybCJez8{d0q~m|YxDc?W+43`6ZV_H`o6 zPuVH5EcF45@>{xS4Bx|dV?W(06~czK+dr)?1~7_X9ZWa+eiTC8s8h?9oQvhz6()V$ zs3G<6o#acIGhQQra=r`7SMp0^L#|a>zrfMP&qpsKCnS?IH={NA&txZ&->jQ|SgA;U zR6L*rW+~X=wh_>X!=*-ZgAbc^NllPpk=4`G}oge9E+)!?E?S zVg42qH|7=(@txoUN?FjQ;=?gH^E6Im4@AD^kd~{> zqdGduOAL2pK3h_Th?T#VWKBP3EeWP~n3IpCNy>c-mYLToh7kiJX~tx&6UjJ6eWn6y zCF!|l!2OugQ+G%32w$SDV_)d--kOMH4|iU@8?0|pVdw{SC?wVKOtZb}Y>DebQQ*a6 zqzAFMNBo&Ggf4N`yEPEuw^3Cyd`JB~r7|J?d1c^3;N|_k=Po01@+Bhi?miab3&0{Y zX?$H6>}E)i-*G+xZc$Rzx(xrT}4M!&RR*E@Y8JNM|4-Nnb9-q(-j%PQ+I&b0*Q=LH>*ZF zMteX9zJO$fL(JaWc-54XQ?R_JFfdqLQz!J5w>)94ff~cam2FQ2xes{0MZ0uLDJh!_c-~)|KeK@?sBzFYYsg6FZ6dCcqtkENUsWzH zaccKQ8Y;%+zl_2Isow5t9;7t+%nIi24~?@^=CN++LNStixC#pZEBQrF{gyGAPc zE-5|N^q!WjF>klXloz>GxBV%;&+A=uVvn?^Ilwo?9v89-izV7x0>ZlKi)*Szi_x{j zHN8Z~Mw{NI7Ulp*h$U4DnwS9nu{BG5vr&mXZgU%_mXAZ&soxs8L7ky5HaU4BA0nn` zN|QCW_08z2Wy;ccTg&dCh{}(Z%hBon_740$2Ib%E!J6Z*bu|_IaPbVHS@%{A{5jAc zq;G~*^}hbbElFJSi9O?xi?0mVeSr4{^JzFBwqcg0hJPq9%(6t~Fxtyx!25pYd^FUN z>K)TRX^r&FECW>gkOSBt+O_|c6Y}z=pBcUN7VZW9m0Mlgm@+yXfRLe~h)rbsnN3M> zN)m7-i{T!OBZCUA8>)(qM~i`5iTAXQ+g9;3+Lf9Y%TWC!_L~9{((_bt$rEu&uhmNT zQHwJpXJKAuelfPl>bi)RWr~e$Tz88TkU014KIdDQrRVL{HeWA%c8GGFR;}sl8!c6H zTligU2!+youR5V8MCplhx6^6=m{XsA%jXNx|E3xDWWnAgfWOSgJdd<$%@Nl5jH^R@ zc50vL@=L{$&)lAaN$e%@oY7&ahFKWlXp7Bb5swyoT)rRwYjPDj`Xpj8J^h=6tJ`JK zY4-u0_JS?WK3qz%uwGko(((W>NojDR!eg+_?ozI{Hev_UHCmY3I!dqlvPcbMwA$D$ zOa~>T8+?jQ?YvIE{qAQrosr&k5kQvCLTgA&JOx}{^A!YDUr=*ld36E)XKtduz?lX1 ziB%igjGz1bP-fI-#~ZPgZ2D-HU098zTanYZ;{BK$nmIFMkort4R^fTI5@$M#d!ImeQGdsq$cgB4kp5NTiVz zlZfp!_k}oqNLE;-U1_tq?hi^2t8IHn6d&_xQx|r8LvWv?Ns=U`bV!o;eWkw*J!LWn z;L}`tMp)kZWK$7kTOO{LAlT+>XPTHM<9IbD{i#A_1GlhC=&8duATX;xBTc}Lm=n=o{V z#OEZo0Tasecm388b$t08(U!|wr|>4x3;9Kz5!%I1e!PvU}Dqo extends Learning imple protected int episodeSumCurrentSecond; protected double sumOfRewards; protected List> episode = new ArrayList<>(); - + protected int timestampCurrentEpisode = 0; + protected boolean converged; public EpisodicLearning(Environment environment, DiscreteActionSpace actionSpace, float discountFactor, int delay) { super(environment, actionSpace, discountFactor, delay); initBenchMarking(); @@ -50,7 +55,7 @@ public abstract class EpisodicLearning extends Learning imple private void initBenchMarking(){ new Thread(()->{ - while (true){ + while (currentlyLearning){ episodePerSecond = episodeSumCurrentSecond; episodeSumCurrentSecond = 0; try { @@ -62,7 +67,7 @@ public abstract class EpisodicLearning extends Learning imple }).start(); } - protected void dispatchEpisodeEnd(){ + private void dispatchEpisodeEnd(){ ++episodeSumCurrentSecond; if(rewardHistory.size() > 10000){ rewardHistory.clear(); @@ -73,20 +78,20 @@ public abstract class EpisodicLearning extends Learning imple } } - protected void dispatchEpisodeStart(){ + private void dispatchEpisodeStart(){ ++currentEpisode; /* - 2f 0.02 => 100 - 1.5f 0.02 => 75 - 1.4f 0.02 => fail - 1.5f 0.1 => 16 ! - */ - if(this.policy instanceof EpsilonGreedyPolicy){ - float ep = 1.5f/(float)currentEpisode; - if(ep < 0.10) ep = 0; - ((EpsilonGreedyPolicy) this.policy).setEpsilon(ep); - System.out.println(ep); - } + 2f 0.02 => 100 + 1.5f 0.02 => 75 + 1.4f 0.02 => fail + 1.5f 0.1 => 16 ! + */ +// if(this.policy instanceof EpsilonGreedyPolicy){ +// float ep = 2f/(float)currentEpisode; +// if(ep < 0.02) ep = 0; +// ((EpsilonGreedyPolicy) this.policy).setEpsilon(ep); +// System.out.println(ep); +// } episodesToLearn.decrementAndGet(); for(LearningListener l: learningListeners){ l.onEpisodeStart(); @@ -97,10 +102,20 @@ public abstract class EpisodicLearning extends Learning imple protected void dispatchStepEnd() { super.dispatchStepEnd(); timestamp++; + timestampCurrentEpisode++; // TODO: more sophisticated way to check convergence - if(timestamp > 300000){ - System.out.println("converged after: " + currentEpisode + " episode!"); - interruptLearning(); + if(timestampCurrentEpisode > 300000){ + converged = true; + // t + File file = new File("convergence.txt"); + try { + Files.writeString(Path.of(file.getPath()), currentEpisode/2 + ",", StandardOpenOption.APPEND); + } catch (IOException e) { + e.printStackTrace(); + } + System.out.println("converged after: " + currentEpisode/2 + " episode!"); + episodesToLearn.set(0); + dispatchLearningEnd(); } } @@ -110,20 +125,23 @@ public abstract class EpisodicLearning extends Learning imple } private void startLearning(){ - learningExecutor.submit(()->{ - dispatchLearningStart(); - while(episodesToLearn.get() > 0){ - dispatchEpisodeStart(); - nextEpisode(); - dispatchEpisodeEnd(); - } - synchronized (this){ - dispatchLearningEnd(); - notifyAll(); - } - }); + dispatchLearningStart(); + while(episodesToLearn.get() > 0){ + + dispatchEpisodeStart(); + timestampCurrentEpisode = 0; + nextEpisode(); + dispatchEpisodeEnd(); + } + synchronized (this){ + dispatchLearningEnd(); + notifyAll(); + } } + public void learnMoreEpisodes(int nrOfEpisodes){ + episodesToLearn.addAndGet(nrOfEpisodes); + } /** * Stopping the while loop by setting episodesToLearn to 0. * The current episode can not be interrupted, so the sleep delay diff --git a/src/main/java/core/algo/Learning.java b/src/main/java/core/algo/Learning.java index fbba9ef..1bb5207 100644 --- a/src/main/java/core/algo/Learning.java +++ b/src/main/java/core/algo/Learning.java @@ -42,8 +42,7 @@ public abstract class Learning{ @Setter protected int delay; protected List rewardHistory; - protected ExecutorService learningExecutor; - protected boolean currentlyLearning; + protected volatile boolean currentlyLearning; public Learning(Environment environment, DiscreteActionSpace actionSpace, float discountFactor, int delay) { this.environment = environment; @@ -53,7 +52,6 @@ public abstract class Learning{ currentlyLearning = false; learningListeners = new HashSet<>(); rewardHistory = new CopyOnWriteArrayList<>(); - learningExecutor = Executors.newSingleThreadExecutor(); } public Learning(Environment environment, DiscreteActionSpace actionSpace, float discountFactor) { @@ -89,8 +87,6 @@ public abstract class Learning{ protected void dispatchLearningEnd() { currentlyLearning = false; - System.out.println("Checksum: " + checkSum); - System.out.println("Reward Checksum: " + rewardCheckSum); for (LearningListener l : learningListeners) { l.onLearningEnd(); } diff --git a/src/main/java/core/algo/mc/MonteCarloControlFirstVisitEGreedy.java b/src/main/java/core/algo/mc/MonteCarloControlFirstVisitEGreedy.java index 50e9ce2..5bc93d3 100644 --- a/src/main/java/core/algo/mc/MonteCarloControlFirstVisitEGreedy.java +++ b/src/main/java/core/algo/mc/MonteCarloControlFirstVisitEGreedy.java @@ -3,12 +3,17 @@ package core.algo.mc; import core.*; import core.algo.EpisodicLearning; import core.policy.EpsilonGreedyPolicy; +import core.policy.GreedyPolicy; +import core.policy.Policy; import org.apache.commons.lang3.tuple.ImmutablePair; import org.apache.commons.lang3.tuple.Pair; -import java.io.IOException; -import java.io.ObjectInputStream; -import java.io.ObjectOutputStream; +import java.io.*; +import java.net.URI; +import java.nio.file.Files; +import java.nio.file.Path; +import java.nio.file.Paths; +import java.nio.file.StandardOpenOption; import java.util.*; /** @@ -35,8 +40,16 @@ public class MonteCarloControlFirstVisitEGreedy extends Episodic private Map, Double> returnSum; private Map, Integer> returnCount; + // t + private float epsilon; + // t + private Policy greedyPolicy = new GreedyPolicy<>(); + + public MonteCarloControlFirstVisitEGreedy(Environment environment, DiscreteActionSpace actionSpace, float discountFactor, float epsilon, int delay) { super(environment, actionSpace, discountFactor, delay); + // t + this.epsilon = epsilon; this.policy = new EpsilonGreedyPolicy<>(epsilon); this.stateActionTable = new DeterministicStateActionTable<>(this.actionSpace); returnSum = new HashMap<>(); @@ -58,12 +71,16 @@ public class MonteCarloControlFirstVisitEGreedy extends Episodic } sumOfRewards = 0; StepResultEnvironment envResult = null; - //TODO extract to learning - int timestamp = 0; + while(envResult == null || !envResult.isDone()) { Map actionValues = stateActionTable.getActionValues(state); - A chosenAction = policy.chooseAction(actionValues); - checkSum += chosenAction.ordinal(); + A chosenAction; + if(currentEpisode % 2 == 1){ + chosenAction = greedyPolicy.chooseAction(actionValues); + }else{ + chosenAction = policy.chooseAction(actionValues); + } + envResult = environment.step(chosenAction); State nextState = envResult.getState(); sumOfRewards += envResult.getReward(); @@ -79,6 +96,11 @@ public class MonteCarloControlFirstVisitEGreedy extends Episodic } timestamp++; dispatchStepEnd(); + if(converged) return; + } + + if(currentEpisode % 2 == 1){ + return; } // System.out.printf("Episode %d \t Reward: %f \n", currentEpisode, sumOfRewards); diff --git a/src/main/java/core/controller/RLController.java b/src/main/java/core/controller/RLController.java index d855ab1..fc33239 100644 --- a/src/main/java/core/controller/RLController.java +++ b/src/main/java/core/controller/RLController.java @@ -67,17 +67,17 @@ public class RLController implements LearningListener { } protected void initListeners() { - learning.addListener(this); - new Thread(() -> { - while(true) { - printNextEpisode = true; - try { - Thread.sleep(30 * 1000); - } catch (InterruptedException e) { - e.printStackTrace(); + learning.addListener(this); + new Thread(() -> { + while(learning.isCurrentlyLearning()) { + printNextEpisode = true; + try { + Thread.sleep(30 * 1000); + } catch (InterruptedException e) { + e.printStackTrace(); + } } - } - }).start(); + }).start(); } private void initLearning() { @@ -95,7 +95,13 @@ public class RLController implements LearningListener { protected void learnMoreEpisodes(int nrOfEpisodes) { if(learning instanceof EpisodicLearning) { - ((EpisodicLearning) learning).learn(nrOfEpisodes); + if(learning.isCurrentlyLearning()){ + ((EpisodicLearning) learning).learnMoreEpisodes(nrOfEpisodes); + }else{ + new Thread(() -> { + ((EpisodicLearning) learning).learn(nrOfEpisodes); + }).start(); + } } else { throw new RuntimeException("Triggering onLearnMoreEpisodes on non-episodic learning!"); } diff --git a/src/main/java/example/JumpingDino.java b/src/main/java/example/JumpingDino.java index 41d5290..13efb4e 100644 --- a/src/main/java/example/JumpingDino.java +++ b/src/main/java/example/JumpingDino.java @@ -3,23 +3,49 @@ package example; import core.RNG; import core.algo.Method; import core.controller.RLController; +import core.controller.RLControllerGUI; import evironment.jumpingDino.DinoAction; import evironment.jumpingDino.DinoWorld; +import java.io.File; +import java.io.IOException; +import java.nio.file.Files; +import java.nio.file.Path; +import java.nio.file.StandardOpenOption; + public class JumpingDino { public static void main(String[] args) { - RNG.setSeed(55); + File file = new File("convergence.txt"); + for(float f = 0.05f; f <=1.003 ; f+=0.05f){ + try { + Files.writeString(Path.of(file.getPath()), f + ",", StandardOpenOption.APPEND); + } catch (IOException e) { + e.printStackTrace(); + } + for(int i = 1; i <= 100; i++) { + System.out.println("seed: " + i *13); + RNG.setSeed(i *13); - RLController rl = new RLController<>( - new DinoWorld(false, false), - Method.MC_CONTROL_FIRST_VISIT, - DinoAction.values()); + RLController rl = new RLController<>( + new DinoWorld(false, false), + Method.MC_CONTROL_FIRST_VISIT, + DinoAction.values()); - rl.setDelay(0); - rl.setDiscountFactor(1f); - rl.setEpsilon(0.15f); - rl.setLearningRate(1f); - rl.setNrOfEpisodes(400); - rl.start(); + + rl.setDelay(0); + rl.setDiscountFactor(1f); + rl.setEpsilon(f); + rl.setLearningRate(1f); + rl.setNrOfEpisodes(20000); + rl.start(); + + } + try { + Files.writeString(Path.of(file.getPath()), "\n", StandardOpenOption.APPEND); + } catch (IOException e) { + e.printStackTrace(); + } + } + System.out.println("kek"); } } diff --git a/src/main/java/example/Results b/src/main/java/example/Results new file mode 100644 index 0000000..ba42fc7 --- /dev/null +++ b/src/main/java/example/Results @@ -0,0 +1,15 @@ +Method: +Epsilon = k / currentEpisode +set to 0 if Epsilon < b + +k = 1.5 +b = 0.1 => conv. 16 + +k = 1.5 +b = 0.02 => 75 + +k = 1.4 +b = 0.02 => fail + +k = 2.0 +b = 0.02 => conv. 100