big2Game.py

#big 2 class
import enumerateOptions
import gameLogic
import numpy as np
import random
import math
from multiprocessing import Process, Pipe

def convertAvailableActions(availAcs):
    #convert from (1,0,0,1,1...) to (0, -math.inf, -math.inf, 0,0...) etc
    availAcs[np.nonzero(availAcs==0)] = -math.inf
    availAcs[np.nonzero(availAcs==1)] = 0
    return availAcs

class handPlayed:
    def __init__(self, hand, player):
        self.hand = hand
        self.player = player
        self.nCards = len(hand)
        if self.nCards <= 3:
            self.type = 1
        elif self.nCards == 4:
            if gameLogic.isFourOfAKind(hand):
                self.type = 2
            else:
                self.type = 1
        elif self.nCards == 5:
            if gameLogic.isStraight(hand):
                if gameLogic.isFlush(hand):
                    self.type = 4
                else:
                    self.type = 1
            elif gameLogic.isFlush(hand):
                self.type = 2
            else:
                self.type = 3

class big2Game:
    def __init__(self):
        self.reset()
        
    def reset(self):
        shuffledDeck = np.random.permutation(52) + 1
        #hand out cards to each player
        self.currentHands = {}
        self.currentHands[1] = np.sort(shuffledDeck[0:13])
        self.currentHands[2] = np.sort(shuffledDeck[13:26])
        self.currentHands[3] = np.sort(shuffledDeck[26:39])
        self.currentHands[4] = np.sort(shuffledDeck[39:52])
        self.cardsPlayed = np.zeros((4,52), dtype=int)
        #who has 3D - this gets played
        for i in range(52):
            if shuffledDeck[i] == 1:
                threeDiamondInd = i
                break
        if threeDiamondInd < 13:
            whoHas3D = 1
        elif threeDiamondInd < 26:
            whoHas3D = 2
        elif threeDiamondInd < 39:
            whoHas3D = 3
        else:
            whoHas3D = 4
        self.currentHands[whoHas3D] = self.currentHands[whoHas3D][1:]
        self.cardsPlayed[whoHas3D-1][0] = 1
        self.goIndex = 1
        self.handsPlayed = {}
        self.handsPlayed[self.goIndex] = handPlayed([1],whoHas3D)
        self.goIndex += 1
        self.playersGo = whoHas3D + 1
        if self.playersGo == 5:
            self.playersGo = 1
        self.passCount = 0
        self.control = 0
        self.neuralNetworkInputs = {}
        self.neuralNetworkInputs[1] = np.zeros((412,), dtype=int)
        self.neuralNetworkInputs[2] = np.zeros((412,), dtype=int)
        self.neuralNetworkInputs[3] = np.zeros((412,), dtype=int)
        self.neuralNetworkInputs[4] = np.zeros((412,), dtype=int)
        nPlayerInd = 22*13
        nnPlayerInd = nPlayerInd + 27
        nnnPlayerInd = nnPlayerInd + 27
        #initialize number of cards
        for i in range(1,5):
            self.neuralNetworkInputs[i][nPlayerInd+12]=1
            self.neuralNetworkInputs[i][nnPlayerInd+12]=1
            self.neuralNetworkInputs[i][nnnPlayerInd+12]=1
        self.fillNeuralNetworkHand(1)
        self.fillNeuralNetworkHand(2)
        self.fillNeuralNetworkHand(3)
        self.fillNeuralNetworkHand(4)
        self.updateNeuralNetworkInputs(np.array([1]),whoHas3D)
        self.gameOver = 0
        self.rewards = np.zeros((4,))
        self.goCounter = 0
        
    def fillNeuralNetworkHand(self,player):
        handOptions = gameLogic.handsAvailable(self.currentHands[player])
        sInd = 0
        self.neuralNetworkInputs[player][sInd:22*13] = 0
        for i in range(len(self.currentHands[player])):
            value = handOptions.cards[self.currentHands[player][i]].value
            suit = handOptions.cards[self.currentHands[player][i]].suit
            self.neuralNetworkInputs[player][sInd+int(value)-1] = 1
            if suit == 1:
                self.neuralNetworkInputs[player][sInd+13] = 1
            elif suit == 2:
                self.neuralNetworkInputs[player][sInd+14] = 1
            elif suit == 3:
                self.neuralNetworkInputs[player][sInd+15] = 1
            else:
                self.neuralNetworkInputs[player][sInd+16] = 1
            if handOptions.cards[self.currentHands[player][i]].inPair:
                self.neuralNetworkInputs[player][sInd+17] = 1
            if handOptions.cards[self.currentHands[player][i]].inThreeOfAKind:
                self.neuralNetworkInputs[player][sInd+18] = 1
            if handOptions.cards[self.currentHands[player][i]].inFourOfAKind:
                self.neuralNetworkInputs[player][sInd+19] = 1
            if handOptions.cards[self.currentHands[player][i]].inStraight:
                self.neuralNetworkInputs[player][sInd+20] = 1
            if handOptions.cards[self.currentHands[player][i]].inFlush:
                self.neuralNetworkInputs[player][sInd+21] = 1
            sInd += 22
    
    def updateNeuralNetworkPass(self, cPlayer):
        #this is a bit of a mess tbh, some things are unnecessary.
        phInd = 22*13 + 27 + 27 + 27 + 16
        nPlayer = cPlayer-1
        if nPlayer == 0:
            nPlayer = 4
        nnPlayer = nPlayer - 1
        if nnPlayer == 0:
            nnPlayer = 4
        nnnPlayer = nnPlayer - 1
        if nnnPlayer == 0:
            nnnPlayer = 4
        if self.passCount < 2:
            #no control - prev hands remain same
            self.neuralNetworkInputs[nPlayer][phInd+26:] = 0
            self.neuralNetworkInputs[nnPlayer][phInd+26:] = 0
            self.neuralNetworkInputs[nnnPlayer][phInd+26:] = 0
            if self.passCount == 0:
                self.neuralNetworkInputs[nPlayer][phInd+27] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+27] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+27] = 1
            else:
                self.neuralNetworkInputs[nPlayer][phInd+28] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+28] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+28] = 1
        else:
            #next player is gaining control.
            self.neuralNetworkInputs[nPlayer][phInd:] = 0
            self.neuralNetworkInputs[nnPlayer][phInd:] = 0
            self.neuralNetworkInputs[nnnPlayer][phInd:] = 0
            self.neuralNetworkInputs[nnnPlayer][phInd+17] = 1
    
    def updateNeuralNetworkInputs(self,prevHand, cPlayer):
        self.fillNeuralNetworkHand(cPlayer)
        nPlayer = cPlayer-1
        if nPlayer == 0:
            nPlayer = 4
        nnPlayer = nPlayer - 1
        if nnPlayer == 0:
            nnPlayer = 4
        nnnPlayer = nnPlayer - 1
        if nnnPlayer == 0:
            nnnPlayer = 4
        nCards = self.currentHands[cPlayer].size
        cardsOfNote = np.intersect1d(prevHand, np.arange(45,53))
        nPlayerInd = 22*13
        nnPlayerInd = nPlayerInd + 27
        nnnPlayerInd = nnPlayerInd + 27
        #next player
        self.neuralNetworkInputs[nPlayer][nPlayerInd:(nPlayerInd+13)] = 0
        self.neuralNetworkInputs[nPlayer][nPlayerInd+nCards-1] = 1 #number of cards
        #next next player
        self.neuralNetworkInputs[nnPlayer][nnPlayerInd:(nnPlayerInd+13)] = 0
        self.neuralNetworkInputs[nnPlayer][nnPlayerInd + nCards-1] = 1
        #next next next player
        self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd:(nnnPlayerInd+13)] = 0
        self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd + nCards-1] = 1
        for val in cardsOfNote:
            self.neuralNetworkInputs[nPlayer][nPlayerInd+13+(val-45)] = 1
            self.neuralNetworkInputs[nnPlayer][nnPlayerInd+13+(val-45)] = 1
            self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+13+(val-45)] = 1
        #prevHand
        phInd = nnnPlayerInd + 27 + 16
        self.neuralNetworkInputs[nPlayer][phInd:] = 0
        self.neuralNetworkInputs[nnPlayer][phInd:] = 0
        self.neuralNetworkInputs[nnnPlayer][phInd:] = 0
        self.neuralNetworkInputs[cPlayer][phInd:] = 0
        nCards = prevHand.size
        
        if nCards == 2:
            self.neuralNetworkInputs[nPlayer][nPlayerInd+21] = 1
            self.neuralNetworkInputs[nnPlayer][nnPlayerInd+21] = 1
            self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+21] = 1
            value = int(gameLogic.cardValue(prevHand[1]))
            suit = prevHand[1] % 4
            self.neuralNetworkInputs[nPlayer][phInd+19] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+19] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+19] = 1
        elif nCards == 3:
            self.neuralNetworkInputs[nPlayer][nPlayerInd+22] = 1
            self.neuralNetworkInputs[nnPlayer][nnPlayerInd+22] = 1
            self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+22] = 1
            value = int(gameLogic.cardValue(prevHand[2]))
            suit = prevHand[2] % 4
            self.neuralNetworkInputs[nPlayer][phInd+20] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+20] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+20] = 1
        elif nCards == 4:
            self.neuralNetworkInputs[nPlayer][nPlayerInd+23] = 1
            self.neuralNetworkInputs[nnPlayer][nnPlayerInd+23] = 1
            self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+23] = 1
            value = int(gameLogic.cardValue(prevHand[3]))
            suit = prevHand[3] % 4
            if gameLogic.isTwoPair(prevHand):
                self.neuralNetworkInputs[nPlayer][phInd+21] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+21] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+21] = 1
            else:
                self.neuralNetworkInputs[nPlayer][phInd+22] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+22] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+22] = 1
        elif nCards == 5:
            #import pdb; pdb.set_trace()
            if gameLogic.isStraight(prevHand):
                self.neuralNetworkInputs[nPlayer][nPlayerInd+24] = 1
                self.neuralNetworkInputs[nnPlayer][nnPlayerInd+24] = 1
                self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+24] = 1
                value = int(gameLogic.cardValue(prevHand[4]))
                suit = prevHand[4] % 4
                self.neuralNetworkInputs[nPlayer][phInd+23] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+23] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+23] = 1
            if gameLogic.isFlush(prevHand):
                self.neuralNetworkInputs[nPlayer][nPlayerInd + 25] = 1
                self.neuralNetworkInputs[nnPlayer][nnPlayerInd + 25] = 1
                self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd+25] = 1
                value = int(gameLogic.cardValue(prevHand[4]))
                suit = prevHand[4] % 4
                self.neuralNetworkInputs[nPlayer][phInd+24] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+24] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+24] = 1
            elif gameLogic.isFullHouse(prevHand):
                self.neuralNetworkInputs[nPlayer][nPlayerInd + 26] = 1
                self.neuralNetworkInputs[nnPlayer][nnPlayerInd + 26] = 1
                self.neuralNetworkInputs[nnnPlayer][nnnPlayerInd + 26] = 1
                value = int(gameLogic.cardValue(prevHand[2]))
                suit = -1
                self.neuralNetworkInputs[nPlayer][phInd+25] = 1
                self.neuralNetworkInputs[nnPlayer][phInd+25] = 1
                self.neuralNetworkInputs[nnnPlayer][phInd+25] = 1
        else:
            value = int(gameLogic.cardValue(prevHand[0]))
            suit = prevHand[0] % 4
            self.neuralNetworkInputs[nPlayer][phInd+18] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+18] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+18] = 1
        self.neuralNetworkInputs[nPlayer][phInd+value-1] = 1
        self.neuralNetworkInputs[nnPlayer][phInd+value-1] = 1
        self.neuralNetworkInputs[nnnPlayer][phInd+value-1] = 1
        if suit == 1:
            self.neuralNetworkInputs[nPlayer][phInd+13] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+13] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+13] = 1
        elif suit == 2:
            self.neuralNetworkInputs[nPlayer][phInd+14] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+14] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+14] = 1
        elif suit == 3:
            self.neuralNetworkInputs[nPlayer][phInd+15] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+15] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+15] = 1
        elif suit == 0:
            self.neuralNetworkInputs[nPlayer][phInd+16] = 1
            self.neuralNetworkInputs[nnPlayer][phInd+16] = 1
            self.neuralNetworkInputs[nnnPlayer][phInd+16] = 1
        #general - common to all hands.
        cardsRecord = np.intersect1d(prevHand, np.arange(37,53))
        endInd = nnnPlayerInd + 27
        for val in cardsRecord:
            self.neuralNetworkInputs[1][endInd+(val-37)] = 1
            self.neuralNetworkInputs[2][endInd+(val-37)] = 1
            self.neuralNetworkInputs[3][endInd+(val-37)] = 1
            self.neuralNetworkInputs[4][endInd+(val-37)] = 1
        #no passes.
        self.neuralNetworkInputs[nPlayer][phInd+26] = 1
        self.neuralNetworkInputs[nnPlayer][phInd+26] = 1
        self.neuralNetworkInputs[nnnPlayer][phInd+26] = 1
        self.neuralNetworkInputs[nPlayer][phInd+27:] = 0
        self.neuralNetworkInputs[nnPlayer][phInd+27:] = 0
        self.neuralNetworkInputs[nnnPlayer][phInd+27:] = 0                
    
    def updateGame(self, option, nCards=0):
        self.goCounter += 1
        if option == -1:
            #they pass
            cPlayer = self.playersGo
            self.updateNeuralNetworkPass(cPlayer)
            self.playersGo += 1
            if self.playersGo == 5:
                self.playersGo = 1
            self.passCount += 1
            if self.passCount == 3:
                self.control = 1
                self.passCount = 0
            return
        self.passCount = 0
        if nCards == 1:
            handToPlay = np.array([self.currentHands[self.playersGo][option]])
        elif nCards == 2:
            handToPlay = self.currentHands[self.playersGo][enumerateOptions.inverseTwoCardIndices[option]]
        elif nCards == 3:
            handToPlay = self.currentHands[self.playersGo][enumerateOptions.inverseThreeCardIndices[option]]
        elif nCards == 4:
            handToPlay = self.currentHands[self.playersGo][enumerateOptions.inverseFourCardIndices[option]]
        else:
            handToPlay = self.currentHands[self.playersGo][enumerateOptions.inverseFiveCardIndices[option]]
        for i in handToPlay:
            self.cardsPlayed[self.playersGo-1][i-1] = 1
        self.handsPlayed[self.goIndex] = handPlayed(handToPlay, self.playersGo)
        self.control = 0
        self.goIndex += 1
        self.currentHands[self.playersGo] = np.setdiff1d(self.currentHands[self.playersGo],handToPlay)
        if self.currentHands[self.playersGo].size == 0:
            self.assignRewards()
            self.gameOver = 1
            return
        self.updateNeuralNetworkInputs(handToPlay, self.playersGo)
        self.playersGo += 1
        if self.playersGo == 5:
            self.playersGo = 1
            
    def assignRewards(self):
        totCardsLeft = 0
        for i in range(1,5):
            nC = self.currentHands[i].size
            if nC == 0:
                winner = i
            else:
                self.rewards[i-1] = -1*nC
                totCardsLeft += nC
        self.rewards[winner-1] = totCardsLeft
        
    def randomOption(self):
        cHand = self.currentHands[self.playersGo]
        if self.control == 0:
            prevHand = self.handsPlayed[self.goIndex-1].hand
            nCards = len(prevHand)
            if nCards > 1:
                handOptions = gameLogic.handsAvailable(cHand)
            if nCards == 1:
                options = enumerateOptions.oneCardOptions(cHand,prevHand,1)
            elif nCards == 2:
                options = enumerateOptions.twoCardOptions(handOptions, prevHand, 1)
            elif nCards == 3:
                options = enumerateOptions.threeCardOptions(handOptions, prevHand, 1)
            elif nCards == 4:
                if gameLogic.isFourOfAKind(prevHand):
                    options = enumerateOptions.fourCardOptions(handOptions, prevHand, 2)
                else:
                    options = enumerateOptions.fourCardOptions(handOptions, prevHand, 1)
            else:
                if gameLogic.isStraight(prevHand):
                    if gameLogic.isFlush(prevHand):
                        options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 4)
                    else:
                        options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 1)
                elif gameLogic.isFlush(prevHand):
                    options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 2)
                else:
                    options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 3)
            if isinstance(options,int):
                nOptions = -1
            else:
                nOptions = len(options)
            ind = random.randint(0,nOptions)
            if ind == nOptions or isinstance(options,int):
                return -1 #pass
            else:
                return (options[ind], nCards)
        else:
            #we have control - choose from any option
            handOptions = gameLogic.handsAvailable(cHand)
            oneCardOptions = enumerateOptions.oneCardOptions(cHand)
            twoCardOptions = enumerateOptions.twoCardOptions(handOptions)
            threeCardOptions = enumerateOptions.threeCardOptions(handOptions)
            fourCardOptions = enumerateOptions.fourCardOptions(handOptions)
            fiveCardOptions = enumerateOptions.fiveCardOptions(handOptions)
            if isinstance(oneCardOptions, int):
                n1 = 0
            else:
                n1 = len(oneCardOptions)
            if isinstance(twoCardOptions, int):
                n2 = 0
            else:
                n2 = len(twoCardOptions)
            if isinstance(threeCardOptions, int):
                n3 = 0
            else:
                n3 = len(threeCardOptions)
            if isinstance(fourCardOptions, int):
                n4 = 0
            else:
                n4 = len(fourCardOptions)
            if isinstance(fiveCardOptions, int):
                n5 = 0
            else:
                n5 = len(fiveCardOptions)
            nTot = n1 + n2 + n3 + n4 + n5
            ind = random.randint(0,nTot-1)
            if ind < n1:
                return (oneCardOptions[ind],1)
            elif ind < (n1+n2):
                return (twoCardOptions[ind-n1],2)
            elif ind < (n1+n2+n3):
                return (threeCardOptions[ind-n1-n2],3)
            elif ind < (n1+n2+n3+n4):
                return (fourCardOptions[ind-n1-n2-n3],4)
            else:
                return (fiveCardOptions[ind-n1-n2-n3-n4],5)
            
    def returnAvailableActions(self):
    
        currHand = self.currentHands[self.playersGo]
        availableActions = np.zeros((enumerateOptions.nActions[5]+1,))
        
        if self.control == 0:
            #allow pass action
            availableActions[enumerateOptions.passInd] = 1
            
            prevHand = self.handsPlayed[self.goIndex-1].hand
            nCardsToBeat = len(prevHand)
            
            if nCardsToBeat > 1:
                handOptions = gameLogic.handsAvailable(currHand)
                
            if nCardsToBeat == 1:
                options = enumerateOptions.oneCardOptions(currHand, prevHand,1)
            elif nCardsToBeat == 2:
                options = enumerateOptions.twoCardOptions(handOptions, prevHand, 1)
            elif nCardsToBeat == 3:
                options = enumerateOptions.threeCardOptions(handOptions, prevHand, 1)
            elif nCardsToBeat == 4:
                if gameLogic.isFourOfAKind(prevHand):
                    options = enumerateOptions.fourCardOptions(handOptions, prevHand, 2)
                else:
                    options = enumerateOptions.fourCardOptions(handOptions, prevHand, 1)
            else:
                if gameLogic.isStraight(prevHand):
                    if gameLogic.isFlush(prevHand):
                        options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 4)
                    else:
                        options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 1)
                elif gameLogic.isFlush(prevHand):
                    options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 2)
                else:
                    options = enumerateOptions.fiveCardOptions(handOptions, prevHand, 3)
                    
            if isinstance(options, int): #no options - must pass
                return availableActions
            
            for option in options:
                index = enumerateOptions.getIndex(option, nCardsToBeat)
                availableActions[index] = 1
                
            return availableActions
        
        
        else: #player has control.
            handOptions = gameLogic.handsAvailable(currHand)
            oneCardOptions = enumerateOptions.oneCardOptions(currHand)
            twoCardOptions = enumerateOptions.twoCardOptions(handOptions)
            threeCardOptions = enumerateOptions.threeCardOptions(handOptions)
            fourCardOptions = enumerateOptions.fourCardOptions(handOptions)
            fiveCardOptions = enumerateOptions.fiveCardOptions(handOptions)
            
            for option in oneCardOptions:
                index = enumerateOptions.getIndex(option, 1)
                availableActions[index] = 1
                
            if not isinstance(twoCardOptions, int):
                for option in twoCardOptions:
                    index = enumerateOptions.getIndex(option, 2)
                    availableActions[index] = 1
                    
            if not isinstance(threeCardOptions, int):
                for option in threeCardOptions:
                    index = enumerateOptions.getIndex(option, 3)
                    availableActions[index] = 1
                    
            if not isinstance(fourCardOptions, int):
                for option in fourCardOptions:
                    index = enumerateOptions.getIndex(option, 4)
                    availableActions[index] = 1
                    
            if not isinstance(fiveCardOptions, int):
                for option in fiveCardOptions:
                    index = enumerateOptions.getIndex(option, 5)
                    availableActions[index] = 1
                    
            return availableActions

    def step(self, action):
        opt, nC = enumerateOptions.getOptionNC(action)
        self.updateGame(opt, nC)
        if self.gameOver == 0:
            reward = 0
            done = False
            info = None
        else:
            reward = self.rewards
            done = True
            info = {}
            info['numTurns'] = self.goCounter
            info['rewards'] = self.rewards
            #what else is worth monitoring?            
            self.reset()
        return reward, done, info
    
    def getCurrentState(self):
        return self.playersGo, self.neuralNetworkInputs[self.playersGo].reshape(1,412), convertAvailableActions(self.returnAvailableActions()).reshape(1,1695)
        
        
        
#now create a vectorized environment
def worker(remote, parent_remote):
    parent_remote.close()
    game = big2Game()
    while True:
        cmd, data = remote.recv()
        if cmd == 'step':
            reward, done, info = game.step(data)
            remote.send((reward, done, info))
        elif cmd == 'reset':
            game.reset()
            pGo, cState, availAcs = game.getCurrentState()
            remote.send((pGo,cState))
        elif cmd == 'getCurrState':
            pGo, cState, availAcs = game.getCurrentState()
            remote.send((pGo, cState, availAcs))
        elif cmd == 'close':
            remote.close()
            break
        else:
            print("Invalid command sent by remote")
            break
        

class vectorizedBig2Games(object):
    def __init__(self, nGames):
        
        self.waiting = False
        self.closed = False
        self.remotes, self.work_remotes = zip(*[Pipe() for _ in range(nGames)])
        self.ps = [Process(target=worker, args=(work_remote, remote)) for (work_remote, remote) in zip(self.work_remotes, self.remotes)]
        
        for p in self.ps:
            p.daemon = True
            p.start()
        for remote in self.work_remotes:
            remote.close()
            
    def step_async(self, actions):
        for remote, action in zip(self.remotes, actions):
            remote.send(('step', action))
        self.waiting = True
        
    def step_wait(self):
        results = [remote.recv() for remote in self.remotes]
        self.waiting = False
        rewards, dones, infos = zip(*results)
        return rewards, dones, infos
    
    def step(self, actions):
        self.step_async(actions)
        return self.step_wait()
        
    def currStates_async(self):
        for remote in self.remotes:
            remote.send(('getCurrState', None))
        self.waiting = True
        
    def currStates_wait(self):
        results = [remote.recv() for remote in self.remotes]
        self.waiting = False
        pGos, currStates, currAvailAcs = zip(*results)
        return np.stack(pGos), np.stack(currStates), np.stack(currAvailAcs)
    
    def getCurrStates(self):
        self.currStates_async()
        return self.currStates_wait()
    
    def close(self):
        if self.closed:
            return
        if self.waiting:
            for remote in self.remotes:
                remote.recv()
        for remote in self.remotes:
            remote.send(('close', None))
        for p in self.ps:
            p.join()
        self.closed = True