private static double H(int i, GlobalState s, int[] a) { double R = 4; int lipos = 0; //if li>0 int aizer = 0; //if ai=0 double ajzercount = 0; AState aS = (AState)s.getLocalStates()[i]; int li = aS.getL(); if (li > 0) { lipos = 1; } if (a[i] == 0) { aizer = 1; } if (lipos * aizer == 1) { for (int j = 0; j < n; j++) { if (a[j] == 0) { ajzercount++; } } return(R / ajzercount); } else { return(0); } //int c = -2, r = 2, w = -1; //// only valid state action pairs are accepted //int z_i = s.getLocalStates()[i].getZ();//get z_i //int y_i = s.getLocalStates()[i].getY();//get y_i //if (z_i == 0 && a[i] == 1) //{ // return c; //} //else if (z_i == 1 && a[i] == 0 && !gamma(i, s))//stay in finished state if the job not completed //{ // return w; //} //else if (z_i == 1 && a[i] == 1 && gamma(i, s))//go to completed state if a job completed //{ // return r; //} //else //{ // return 0; //} return(0); }
private static GlobalState BT(GlobalState s, int[] a, List <GlobalState> gstates) { State[] localstates = new AState[n]; for (int i = 0; i < n; i++) { AState aS = (AState)(s.getLocalStates()[i]); int label = aS.getLabel(); int l = aS.getL(); int[] c = aS.getC(); int[] aSData = new int[n + 2]; if (label != 2) { int shootcount = 0; for (int j = 0; j < n; j++) { if (j != i && a[j] == (i + 1))//convert zero based index of players to one based index for shooting { shootcount++; if (c[j] > 0)//lose confidence with shootings { c[j]--; } } aSData[j] = c[j]; } l = Math.Max(0, l - shootcount); label = (label + 1) % 3; aSData[n] = l; aSData[n + 1] = label; } else { for (int j = 0; j < n; j++) { if (j != (i))//update confidence { c[j] = Math.Min(2, c[j] + 1); } aSData[j] = c[j]; } l = 2;//restore life label = (label + 1) % 3; aSData[n] = l; aSData[n + 1] = label; } localstates[i] = new AState(i, aSData); } return(gstates[getGlobalStateIndex(localstates)]); }
private static double Pr(int i, GlobalState s, int[] a) { AState aS = (AState)(s.getLocalStates()[i]); //retrive the state of player i int[] c = aS.getC(); //retrive confidence list of player i double normal_sum = 0; double maxcij = 0; for (int j = 0; j < n; j++) { AState aSj = (AState)(s.getLocalStates()[j]); if (j != i) { if (aSj.getL() > 0) //only the living ones are considered for shooting { if (c[j] > maxcij) //find max confidence value { maxcij = c[j]; } normal_sum += c[j]; } } } if (normal_sum > 0)//if some other player is alive { if (a[i] == 0) { return(1 - maxcij / (normal_sum)); } else { for (int j = 1; j <= n; j++) { if (a[i] == j) { AState aSj = (AState)(s.getLocalStates()[j - 1]);//shift the index of j int lj = aSj.getL(); if (lj == 0) { return(0); } else { return((maxcij * c[j - 1]) / normal_sum);//shifting the index to zero based } } } } } else //if no other player is alive { if (a[i] == 0) { return(1); } else { return(0); } } return(0); }
private static String Nashepdeltamul(List <double>[] vcurrent, List <double>[] vprev, List <double>[] ucurrent, List <double>[] uprev, int i0, int j0) { String result = ""; double min_distance = 20, deltastar = 0; List <int> ratlist = new List <int>(); ratlist.Add(i0); ratlist.Add(j0); while (result.Equals("undecided") || result.Equals("")) { //initialize vi and ui for (int i = 0; i < n; i++) { vcurrent[i] = new List <double>(); vprev[i] = new List <double>(); ucurrent[i] = new List <double>(); uprev[i] = new List <double>(); for (int s = 0; s < gstates.Count; s++) { vcurrent[i].Add(0); vprev[i].Add(0); ucurrent[i].Add(0); uprev[i].Add(0); } } //algo double M = 4; //beta^k < (1-beta)*delta/5M //k>log((1-beta)*delta/5M)/log(beta) double k = Math.Ceiling(Math.Log(((1 - beta) * delta) / (24 * M)) / Math.Log(beta)); double EiK = (6 * M * Math.Pow(beta, k)) / (1 - beta); System.Diagnostics.Debug.WriteLine("k =" + k); Parallel.For(0, gstates.Count, index => // Parallel.For(0, 729, index => { GlobalState s = gstates[(index * 128) % gstates.Count]; // GlobalState s = gstates[(index * 128) % 729]; for (int t = 0; t < k; t++) { // foreach (int i in ratlist)//i,j rationals // {//for each altruistic player, player of interest is rational bybool[i0] = true; //make i byzantine bybool[j0] = true; //make j byzantine double max = 0; foreach (GlobalAction a_ij in multiAc)//for each action of rational ij 2 { double minbyzi = 20; double minbyzj = 20; foreach (GlobalAction a_z in byzAc)//for each byz action Z { double rewardi = 0; double rewardj = 0; foreach (GlobalAction a_al in AlmrAc) //for each altruistic action,two rat players Al-2 { setglobalAction(i0, j0, a_ij, a_z, a_al); //call overridden method if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); rewardi += GlobalPr(s, Actions) * (H(i0, s, Actions) + beta * vprev[i0][getGlobalStateIndex(snext.getLocalStates())]); rewardj += GlobalPr(s, Actions) * (H(j0, s, Actions) + beta * vprev[j0][getGlobalStateIndex(snext.getLocalStates())]); } }//expected reward calculated for a_z,a_ij if (AlmrAc.Count == 0) { setglobalAction(i0, j0, a_ij, a_z, null);//call overridden method if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); rewardi = (H(i0, s, Actions) + beta * vprev[i0][getGlobalStateIndex(snext.getLocalStates())]); rewardj = (H(j0, s, Actions) + beta * vprev[j0][getGlobalStateIndex(snext.getLocalStates())]); } } if (rewardi < minbyzi) { minbyzi = rewardi; } if (rewardj < minbyzj) { minbyzj = rewardj; } //min reward calculated for each a_ij // System.Diagnostics.Debug.WriteLine("reward =" + minbyz); }//minbyz calculated if (byZ.Count == 0) { double rewardi = 0; double rewardj = 0; foreach (GlobalAction a_al in AlmrAc) //for each altruistic action,two rat players Al-2 { setglobalAction(i0, j0, a_ij, null, a_al); //call overridden method if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); rewardi += GlobalPr(s, Actions) * (H(i0, s, Actions) + beta * vprev[i0][getGlobalStateIndex(snext.getLocalStates())]); rewardj += GlobalPr(s, Actions) * (H(j0, s, Actions) + beta * vprev[j0][getGlobalStateIndex(snext.getLocalStates())]); } } //expected reward calculated for a_z,a_ij if (AlmrAc.Count == 0) { setglobalAction(i0, j0, a_ij, null, null); //call overridden method if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); rewardi = (H(i0, s, Actions) + beta * vprev[i0][getGlobalStateIndex(snext.getLocalStates())]); rewardj = (H(j0, s, Actions) + beta * vprev[j0][getGlobalStateIndex(snext.getLocalStates())]); } } minbyzi = rewardi; minbyzj = rewardj; //min reward calculated for each a_ij // System.Diagnostics.Debug.WriteLine("reward =" + minbyz); } /* if (minbyz > max) * max = minbyz; * exp += Pr(i, s, Actions) * minbyz;*/ //min reward is used to maximize a different reward function if (alpha * minbyzi + (1 - alpha) * minbyzj > max) { max = alpha * minbyzi + (1 - alpha) * minbyzj; vcurrent[i0][getGlobalStateIndex(s.getLocalStates())] = minbyzi; vcurrent[j0][getGlobalStateIndex(s.getLocalStates())] = minbyzj; } }//maxrat calculated // vcurrent[i0][getGlobalStateIndex(s.getLocalStates())] = max;//v^k_i(s)=maxminE // ucurrent[i][getGlobalStateIndex(s.getLocalStates())] = exp;//v^k_i(s)=maxminE bybool[i0] = false; //remove byzantine role from i bybool[j0] = false; //remove byzantine role from i vprev[i0] = vcurrent[i0]; vprev[j0] = vcurrent[j0]; } }); double[] bigDelta = new double[n]; double ep1, ep2; int countfail = 0; int countpass = 0; double distance = 0; double minxy = 20; bigDelta[ratlist[0]] = vcurrent[ratlist[0]][getGlobalStateIndex(gstates[0].getLocalStates())] - ucurrent[ratlist[0]][getGlobalStateIndex(gstates[0].getLocalStates())]; bigDelta[ratlist[1]] = vcurrent[ratlist[1]][getGlobalStateIndex(gstates[0].getLocalStates())] - ucurrent[ratlist[1]][getGlobalStateIndex(gstates[0].getLocalStates())]; double epi1, epi2, epj1, epj2; epi1 = bigDelta[ratlist[0]] - 2 * EiK; epi2 = bigDelta[ratlist[0]] + 2 * EiK; epj1 = bigDelta[ratlist[1]] - 2 * EiK; epj2 = bigDelta[ratlist[1]] + 2 * EiK; if ((epi1 >= 0 && epj1 > eps) || (epj1 >= 0 && epi1 > eps)) { return("unstable"); } else if (epi2 <= eps && epj2 <= eps) { return("stable with eps" + eps); } else if (epi1 <= eps && epj1 <= eps) { return("stable with epsdel" + (eps + delta));//if (eps <= epi2 || eps <= epj2) } else { distance = Math.Sqrt(Math.Pow(epi2 - eps, 2) + Math.Pow(epj2 - eps, 2)); if (distance < min_distance) { min_distance = distance; deltastar = Math.Min(Math.Abs(epi2 - eps), Math.Abs(epj2 - eps)); } result = "undecided"; } alpha += 0.2; if (alpha >= 1) { return("NEepsdelta*" + " " + deltastar); } } // System.Diagnostics.Debug.WriteLine(result); return(result); }
private static void reward_calc(List <double>[] prev, List <double>[] current, int i, bool biz) { /* Parallel.ForEach(gstates, (s) => * {*/ Parallel.For(0, gstates.Count, index => { if (_debug) { Console.WriteLine(System.Threading.Thread.CurrentThread.ManagedThreadId); } GlobalState s = gstates[(index * 128) % gstates.Count]; //GlobalState s = gstates[(index * 128)%59049]; double max = 0; double exp = 0; foreach (int a_i in localactionset[i])//for each action of rational i { double minbyz = 20; if (byzAc.Count == 0)// if no byzantines { double reward = 0; // double ureward = 0; foreach (GlobalAction a_al in AlrAc)//for each altruistic action Al-i { setglobalAction(i, a_i, a_al); if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); reward += GlobalPr(s, Actions) * (H(i, s, Actions) + beta * prev[i][getGlobalStateIndex(snext.getLocalStates())]); // ureward += GlobalPr(s, Actions) * (H(i, s, Actions) + beta * uprev[i][getGlobalStateIndex(snext.getLocalStates())]); } }//expected reward calculated for a_z,a_i minbyz = reward; } else { foreach (GlobalAction a_z in byzAc)//for each byz action Z { double reward = 0; // double ureward = 0; foreach (GlobalAction a_al in AlrAc)//for each altruistic action Al-i { setglobalAction(i, a_i, a_z, a_al); if (BT(s, Actions, gstates) != null) { GlobalState snext = BT(s, Actions, gstates); reward += GlobalPr(s, Actions) * (H(i, s, Actions) + beta * prev[i][getGlobalStateIndex(snext.getLocalStates())]); // ureward += GlobalPr(s, Actions) * (H(i, s, Actions) + beta * uprev[i][getGlobalStateIndex(snext.getLocalStates())]); } }//expected reward calculated for a_z,a_i if (reward < minbyz) { minbyz = reward; } // System.Diagnostics.Debug.WriteLine("reward =" + minbyz); }//minbyz calculated } if (minbyz > max) { max = minbyz; } exp += Pr(i, s, Actions) * minbyz; }//maxrat calculated if (biz) { current[i][getGlobalStateIndex(s.getLocalStates())] = max;//v^k_i(s)=maxminE } else { current[i][getGlobalStateIndex(s.getLocalStates())] = exp;//v^k_i(s)=maxminE } }); }