def bestPolyFit():
import matplotlib.pyplot as plt
data, headers = para.readFile('data_/3_D_SYSTEMS_CORP_DEL.csv')
dates = data['Date']
close = data['Close']
#groups = grouping.groupUp(data, close)
start = 230
end = 379
#dataList = para.averageLast(5)([close], len(close))
dataList = close
#group = groups[23]
#coeffs = fitPoly(3)(dataList, group[0], group[1])
plt.plot(dataList[start:end])
bestRatio = 0
bestDegree = -1
lastValue = -1
savedPlot = None
for i in range(2,20):
coeffs = fitPoly(i)(dataList, start, end)
p = polynomialFunction(coeffs)
estimate = list(map(p, range(0,end-start)))
score = similarity.lpNorms(2)(estimate, dataList[start:end])
print('Degree ' + str(i) + ' score = ' + str(score))
#plt.plot(estimate)
if lastValue == -1:
lastValue = score
else:
ratio = lastValue/score
lastValue = score
if ratio > bestRatio:
savedPlot = estimate
bestRatio = ratio
bestDegree = i
print('Best degree = ' + str(bestDegree) + ' with score = ' + str(bestRatio))
plt.plot(savedPlot)
plt.show()
def testAlgo(algo, target):
# global data
# testAlgoOnData(algo, target, data['Close'])
#
#def testAlgoOnData(algo, target, dataList):
dates = data['Date']
groups = grouping.groupUp(data, data['Close'])
targetNext = target+const.ma
if targetNext >= len(groups):
return None
similarity._normalizeFuns = [similarity.byMean]
similarity._measureFun = algo
results = compareAllGroupsBefore(groups, target)
results2 = compareAllGroupsBefore(groups, targetNext)
results.reverse()
results.sort(key=lambda x : x[2])
results2.sort(key=lambda x : x[2])
### Uses Average Data: useOnlyAverageData = True
#tradePolicy = tradingmeasure.dontSell
#tradePolicy = tradingmeasure.sellOrKeep
#tradePolicy = tradingmeasure.riskAverseSellOrKeep
#tradePolicy = tradingmeasure.largestReturn
### Doesn't use Average Data: useOnlyAverageData = False
tradePolicy = tradingmeasure.confidenceFilter(0.2, tradingmeasure.sellOrKeep)
useOnlyAverageData = False
totalRank = 0
lpScore = 0
nResults = 10
for v in results[0:nResults]:
rank = getRank(results2, v[0]+const.ma)
totalRank += rank
lpScore += similarity.computeWith(groups[v[0]+const.ma], groups[targetNext], [similarity.byFirst], similarity.lpNorms(2))
dataLists = getDataLists(groups, results[0:nResults], const.ma)
if usingOnlyAverageData:
dataLists = tradingmeasure.averageData(dataLists)
money = tradingmeasure.computeWithFunOn(dataLists, groups[targetNext][2], tradePolicy)
#print(money)
totalRank *= 100 # normalize totalRank for equal weightage.
totalRank /= len(results2) # normalize totalRank for equal weightage.
return (lpScore/nResults, totalRank/nResults, money)
#'frechet': similarity.tsdist('frechetDistance'), # (USE?) prints a lot of nonsense
'inf.norm': similarity.tsdist('inf.normDistance'),
'int.per': similarity.tsdist('int.perDistance'),
'lbKeogh_3': similarity.tsdist('lb.keoghDistance', 3),
'lcss_05': similarity.tsdist('lcssDistance', 0.05),
'lcss_15': similarity.tsdist('lcssDistance', 0.15),
'lcss_30': similarity.tsdist('lcssDistance', 0.3),
'lcss_50': similarity.tsdist('lcssDistance', 0.5),
'lp': similarity.tsdist('lpDistance'),
'manhattan': similarity.tsdist('manhattanDistance'),
'mindist.sax_1': similarity.tsdist('mindist.saxDistance',1),
'mindist.sax_2': similarity.tsdist('mindist.saxDistance',2),
'mindist.sax_4': similarity.tsdist('mindist.saxDistance',4),
'mindist.sax_8': similarity.tsdist('mindist.saxDistance',8),
'mindist.sax_16': similarity.tsdist('mindist.saxDistance',16),
'minkowski_25': similarity.lpNorms(2.5), #otherwise known as lp-norms
'minkowski_30': similarity.lpNorms(3),
'minkowski_05': similarity.lpNorms(0.5),
#'ncd': similarity.tsdist('ncdDistance'), # Unknown internal error
'pacf': similarity.tsdist('pacfDistance'),
'pdc': similarity.tsdist('pdcDistance'),
'per': similarity.tsdist('perDistance'),
#'pred': similarity.tsdist('predDistance'),
#'spec.glk': similarity.tsdist('spec.glkDistance'), # {USE} SLOW. Also, I'm getting strange L-BFGS-B errors.
#'spec.isd': similarity.tsdist('spec.isdDistance'), # {USE) SLOW. Also, I'm getting strange L-BFGS-B errors.
'spec.llr': similarity.tsdist('spec.llrDistance'),
'sts': similarity.tsdist('stsDistance'),
'tquest': similarity.tsdist('tquestDistance', tau=0.5), #seems to do nothing...?
'wav': similarity.tsdist('wavDistance'),
}
def testAlgoWeighted(algo, target, weightDataFun, data):
dates = data['Date']
groupsWeighted = grouping.groupUp(data, weightDataFun(data))
groupsClose = grouping.groupUp(data, data['Close'])
targetNext = target+const.ma
if targetNext >= len(groupsWeighted):
return None
similarity._normalizeFuns = [similarity.byMean]
similarity._measureFun = algo
results = testalgos.compareAllGroupsBefore(groupsWeighted, target)
results2 = testalgos.compareAllGroupsBefore(groupsClose, targetNext)
results.reverse()
results.sort(key=lambda x : x[2])
results2.sort(key=lambda x : x[2])
tradePolicy = tradingmeasure.sellOrKeep
usingOnlyAverageData = True
totalRank = 0
lpScore = 0
nResults = 10
for v in results[0:nResults]:
rank = testalgos.getRank(results2, v[0]+const.ma)
totalRank += rank
lpScore += similarity.computeWith(groupsClose[v[0]+const.ma], groupsClose[targetNext], [similarity.byFirst], similarity.lpNorms(2))
dataLists = testalgos.getDataLists(groupsClose, results[0:nResults], const.ma)
if usingOnlyAverageData:
dataLists = tradingmeasure.averageData(dataLists)
money = tradingmeasure.computeWithFunOn(dataLists, groupsClose[targetNext][2], tradePolicy)
#totalRank *= 100 # normalize totalRank for equal weightage.
#totalRank /= len(results2) # normalize totalRank for equal weightage.
return (lpScore/nResults, totalRank/nResults, money)