def __getNewBCM(data, description=None): BCM = BrownClusterModel() BCM.addTrainingData(data) dataMidpoint = len(data)/2 if description is None: description = ''.join(map(str,BCM.wordSequence[dataMidpoint-5:dataMidpoint+5])) BCM.description = description # Defined Only Here... Kind of sloppy. return BCM
class DrawHandler_BCM:
SEQUENCE_CLUSTER_NUM_KEY = 'cl'
SEQUENCE_DATA_KEY = 'dat'
SEQUENCE_WORD_KEY = 'wd'
def __init__(self, addTestData=True):
self.BCM = BrownClusterModel()
if addTestData:
self.addTestData_1OD()
def addTestData_1OD(self, numAssets=3, dataLength=200, differencePeriod=10, addZeroPrefix=True):
from EMF_DataGenerator_util import findFirstOrderDifferences
from stats_util import CorrelatedReturnHistoryGenerator
self.rawTestData = CorrelatedReturnHistoryGenerator(numAssets=numAssets,
dataLength=dataLength,
addZeroPrefix=addZeroPrefix).history
self.wordTestData = findFirstOrderDifferences(self.rawTestData, differencePeriod)
self.BCM.addTrainingData(self.wordTestData)
self.addZeroPrefix = addZeroPrefix
self.differencePeriod = differencePeriod
def createDrawFriendlyDictionary_10D(self, dataSequence, clusterSequence, wordSequence, differencePeriod):
# Get Sequence Data
(dataLength, dataWidth) = dataSequence.shape
(wordLength, wordWidth) = wordSequence.shape
# Verify Data Shapes
if type(clusterSequence) == list:
assert len(clusterSequence) == wordLength
else:
assert clusterSequence.shape == (wordLength, 1)
clusterSequence = clusterSequence[:,0]
assert wordLength + differencePeriod == dataLength
# Create Sequence Data
wordMin = dataMin = maxint
wordMax = dataMax = -maxint - 1
sequenceData = []
for i in range(dataLength):
d = {}
d[self.SEQUENCE_DATA_KEY] = tuple(dataSequence[i,:])
dataMin = min(dataMin, min(dataSequence[i,:]))
dataMax = max(dataMax, max(dataSequence[i,:]))
if i >= differencePeriod:
j = i - differencePeriod
d[self.SEQUENCE_CLUSTER_NUM_KEY] = clusterSequence[j]
d[self.SEQUENCE_WORD_KEY] = tuple(wordSequence[j,:])
wordMin = min(wordMin, min(wordSequence[j,:]))
wordMax = max(wordMax, max(wordSequence[j,:]))
else:
d[self.SEQUENCE_CLUSTER_NUM_KEY] = None
d[self.SEQUENCE_WORD_KEY] = None
sequenceData.append(d)
# Write Data to Dictionary
drawDict = {}
drawDict['hasDate'] = False
drawDict['sequenceData'] = sequenceData
drawDict['differencePeriod'] = differencePeriod
drawDict['dataStart'] = 0
drawDict['dataEnd'] = dataLength
drawDict['dataWidth'] = dataWidth
drawDict['dataMin'] = dataMin # Deprecable, but makes json smoother
drawDict['dataMax'] = dataMax # Deprecable, but makes json smoother
drawDict['wordStart'] = differencePeriod
drawDict['wordEnd'] = drawDict['dataEnd']
drawDict['wordWidth'] = wordWidth
drawDict['wordMin'] = wordMin # Deprecable, but makes json smoother
drawDict['wordMax'] = wordMax # Deprecable, but makes json smoother
return drawDict
def __wordDisplayPad(self, value):
return value*1.2
def getSequenceData_DrawFriendlyDictionary_1OD(self, drawDict, key, column):
sequenceData = drawDict['sequenceData']
if key == self.SEQUENCE_DATA_KEY:
start = drawDict['dataStart']
end = drawDict['dataEnd']
return [d[key][column] for d in sequenceData[start:end]]
elif key == self.SEQUENCE_WORD_KEY:
start = drawDict['wordStart']
end = drawDict['wordEnd']
return [d[key][column] for d in sequenceData[start:end]]
elif key == self.SEQUENCE_CLUSTER_NUM_KEY:
start = drawDict['wordStart']
end = drawDict['wordEnd']
return [d[key] for d in sequenceData[start:end]]
else:
raise Exception('Key Type not recognized')
def drawBCMUsingPyPlot(self, numberOfLoops=None):
numClusters = len(self.BCM.clusters)
self.colors = self.drawBCMUsingPyPlot_get_N_HexColors(numClusters)
moreToMerge = True
count = 0
while moreToMerge:
moreToMerge = self.BCM.mergeClusters_mergeTop(updateClusterSequence=True)
# TODO: In future, create methods to only update clusterSequence
# Do we want DrawDict to be an object eventually?
drawDictionary = self.createDrawFriendlyDictionary_10D( self.rawTestData,
self.BCM.clusterSequence,
self.wordTestData,
self.differencePeriod)
self.drawBCMUsingPyPlot_singleDrawing(drawDictionary)
count += 1
if numberOfLoops is not None and count >= numberOfLoops:
return drawDictionary
def drawBCMUsingPyPlot_singleDrawing(self, drawDict):
plotStart = min(drawDict['dataStart'], drawDict['wordStart'])
plotEnd = max(drawDict['dataEnd'], drawDict['wordEnd'])
plt.figure(1)
# Draw Raw Data
plt.subplot(211)
plt.xlim(plotStart, plotEnd)
xValsData = range(drawDict['dataStart'], drawDict['dataEnd'])
for data_i in range(drawDict['dataWidth']):
# pu.db #TEST
plt.plot( xValsData,
self.getSequenceData_DrawFriendlyDictionary_1OD(drawDict, self.SEQUENCE_DATA_KEY, data_i))
# Draw Words
subplot = plt.subplot(212)
plt.xlim(plotStart, plotEnd)
plt.ylim(self.__wordDisplayPad(drawDict['wordMin']), self.__wordDisplayPad(drawDict['wordMax']))
xVals_Words = range(drawDict['wordStart'], drawDict['wordEnd'])
for word_i in range(drawDict['wordWidth']):
plt.plot( xVals_Words,
self.getSequenceData_DrawFriendlyDictionary_1OD(drawDict, self.SEQUENCE_WORD_KEY, word_i))
self.drawBCMUsingPyPlot_drawClusterPatches(subplot, drawDict)
# Show Drawing
plt.show()
def drawBCMUsingPyPlot_drawClusterPatches(self, subplot, drawDict):
lowY = self.__wordDisplayPad(drawDict['wordMin'])
height = self.__wordDisplayPad(drawDict['wordMax']) - self.__wordDisplayPad(drawDict['wordMin'])
width = 1.0
for i in range(drawDict['wordStart'], drawDict['wordEnd']):
subplot.add_patch(
patches.Rectangle(
(i, lowY), width, height,
alpha=0.5,
facecolor=self.colors[self.BCM.clusterSequence[i-drawDict['differencePeriod']]],
edgecolor='none'
))
# Taken from online (source forgotten...)
def drawBCMUsingPyPlot_get_N_HexColors(self, N=5):
from colorsys import hsv_to_rgb
HSV_tuples = [(x*1.0/N, 0.5, 0.5) for x in xrange(N)]
hex_out = []
for rgb in HSV_tuples:
rgb = map(lambda x: int(x*255), hsv_to_rgb(*rgb))
hexString = "".join(map(lambda x: chr(x).encode('hex'),rgb))
hex_out.append('#'+hexString)
return hex_out