def rankPop(chromos):
errors, results = [], []
# print chromos
for chromo in chromos:
selectedKeys = []
for i in range(len(chromo)):
if chromo[i] == 1:
selectedKeys.append(keys[i])
selectedKeys.append('"PRICE"')
scores = myownq.runInnerLoop(selectedKeys, myownq.qAgent())
if scores[0] <= 0:
errors.append(1)
else:
errors.append(scores[0])
results.append(scores)
# print scores
# print errors
fitnessScores = calcFitness(errors) # calc fitness scores from the erros calculated
pairedPop = zip(
chromos, errors, results, fitnessScores
) # pair each chromo with its protein, ouput and fitness score
rankedPop = sorted(
pairedPop, key=operator.itemgetter(-1), reverse=True
) # sort the paired pop by ascending fitness score
return rankedPop
stateDef += [j]
break
self.stateDef = tuple(stateDef)
#print (self.stateDef)
def getScore(self):
if (self.newPrice == None or self.curPrice == None):
return 0.0
priceChange = float(self.newPrice - self.curPrice)
return priceChange
#-1 = predict lower price, 1 = predict higher price
def getLegalActions(self):
return [-1,1]
def getDef(self):
return self.stateDef
agent = myownq.qAgent()
#The function that runs the inner Q-learning
def run():
initialized = False
# open data row by row to avoid memory overflow
fname = 'data_cleaned.csv'
with open(fname, 'r+') as f:
# this reads in one line at a time from stdin
date_previous = None
ticker_previous = None
observation = {}
observationPrevious = {}
isFirstObservation = True
lineo = 0
hashVal = hash(tuple(selectedKeys))
if (hashVal not in visited):
repeatedState = False
if (numRepeatedStates >=1000):
STOPIT = True
break
if (STOPIT):
break
print ("FINAL SOLUTION")
print (maxState)
def annealSchedule(delta, time):
return math.exp(float(delta) / float(100001 - time)**0.4)
simulatedAgent = myownq.qAgent()
#The function that runs the inner Q-learning
def runSimulatedAnnealing():
initialized = False
#no price in these keys
keys = ['"ACCOCI"', '"ASSETS"', '"ASSETSC"', '"ASSETSNC"', '"BVPS"', '"CAPEX"', '"CASHNEQ"', '"COR"', '"CURRENTRATIO"', '"DE"', '"DEBT"', '"DEPAMOR"', '"DILUTIONRATIO"', '"DPS"', '"EBIT"', '"EBITDA"', '"EBT"', '"EPS"', '"EPSDIL"', '"EQUITY"', '"FCF"', '"FCFPS"', '"GP"', '"INTANGIBLES"', '"INTEXP"', '"INVENTORY"', '"LIABILITIES"', '"LIABILITIESC"', '"LIABILITIESNC"', '"NCF"', '"NCFCOMMON"', '"NCFDEBT"', '"NCFDIV"', '"NCFF"', '"NCFI"', '"NCFO"', '"NCFX"', '"NETINC"', '"NETINCCMN"', '"NETINCDIS"', '"PAYABLES"', '"PB"', '"PREFDIVIS"', '"RECEIVABLES"', '"RETEARN"', '"REVENUE"', '"RND"', '"SGNA"', '"SHARESWA"', '"SHARESWADIL"', '"TANGIBLES"', '"TAXEXP"', '"TBVPS"', '"WORKINGCAPITAL"']
selectedKeys = random.sample(keys, 10)
selectedKeys += ['"PRICE"']
remainingKeys = [key for key in keys if key not in selectedKeys]
hashVal = hash(tuple(selectedKeys))
fScoreCorrect = dict()
fScoreRewards = dict()
previousState = dict()
visited = []
maxState = dict()
def runSimulatedAnnealing():
initialized = False
#no price in these keys
selectedKeys = random.sample(keys, 10)
selectedKeys += ['"PRICE"']
remainingKeys = [key for key in keys if key not in selectedKeys]
hashVal = hash(tuple(selectedKeys))
fScoreCorrect = dict()
fScoreRewards = dict()
previousState = dict()
visited = []
maxState = dict()
maxAgent = None
testing = False
#run the iterations
for x in range(1,100000):
# if (x % 100 == 0):
# print x
# if (x < 10):
result = myownq.runInnerLoop(selectedKeys, myownq.qAgent())
# else:
# if (not testing):
# #print maxState
# testing = True
# maxAgent = maxState["agent"]
# selectedKeys = maxState["selected"]
# result = myownq.runTestLoop(selectedKeys, maxAgent)
# #print "RESULT"
# #print result
# #print maxState
fScoreRewards[hashVal] = result[0]
fScoreCorrect[hashVal] = result[1]
agent = result[2]
visited.append(hashVal)
move = False
#move if we are on the first step
if (len(previousState.keys()) == 0):
move = True
else:
delta = fScoreRewards[hashVal] - fScoreRewards[previousState["hash"]]
#move if we move to a better state
if (delta >= 0):
move = True
#move with a probablity defined by the anneal schedule function
else:
move = random.random() <= annealSchedule(delta, x)
if (len(previousState.keys()) == 0 or move):
previousState["hash"] = hashVal
previousState["selected"] = selectedKeys
previousState["remaining"] = remainingKeys
# print (fScoreRewards[hashVal])
# print (fScoreCorrect[hashVal])
# print x
if (len(maxState.keys()) == 0 or fScoreRewards[hashVal] > maxState["score"]):
maxState["hash"] = hashVal
maxState["selected"] = selectedKeys
maxState["remaining"] = remainingKeys
maxState["score"] = fScoreRewards[hashVal]
maxState["correct"] = fScoreCorrect[hashVal]
maxState["agent"] = agent
repeatedState= True
numRepeatedStates = 0
#avoid repeating states
while (repeatedState):
numRepeatedStates += 1
hashVal = previousState["hash"]
selectedKeys = previousState["selected"]
remainingKeys = previousState["remaining"]
#either randomly remove a fundamental, or add one
#don't want too few fundamentals, so only remove if there are more than 4
#also don't want too many, so limit it at like 45
if ((len(selectedKeys) > 4 and util.flipCoin(0.5)) or len(selectedKeys) > 45):
randomKey = random.sample(selectedKeys, 1)[0]
while (randomKey == '"PRICE"'):
randomKey = random.sample(selectedKeys, 1)[0]
selectedKeys.remove(randomKey)
remainingKeys.append(randomKey)
else:
randomKey = random.sample(remainingKeys, 1)[0]
selectedKeys.append(randomKey)
remainingKeys.remove(randomKey)
hashVal = hash(tuple(selectedKeys))
if (hashVal not in visited):
repeatedState = False
if (numRepeatedStates >=1000):
selectedKeys = random.sample(keys, 10)
selectedKeys += ['"PRICE"']
remainingKeys = [key for key in keys if key not in selectedKeys]
# print ("FINAL SOLUTION")
print (maxState)