def visualizeTrainedModel_v1():
'''
3d Surface plot
'''
modelName = 'mix-fKDE2_cv'
modelPath = 'models/{0}'.format(modelName)
trainedModels = random.sample(os.listdir(modelPath), 10)
for m in trainedModels:
mFile = '{0}/{1}'.format(modelPath, m)
print(mFile)
pId = mFile.split('/')[-1].replace('.model', '').replace(
'_{0}'.format(modelName), '')
print(pId)
m = comparemethods.mixfKDEModel(pId, modelFile=mFile)
m.loadModel()
kde = m.trainedModel
# xs: advanced day
# ys: price kpi
# fixed: seat class, daytype
daytype = commonoperation.DayType.WORKDAY
hourtype = 15 / 2
seat = 1
discount = 0.8
fig = plt.figure()
ax = fig.gca(projection='3d')
X = np.linspace(0, 30, 61) #num of advanced days
Y = np.linspace(0, 1, 51) #price kpi
X, Y = np.meshgrid(X, Y)
old_shape = X.shape
N = X.shape[0] * X.shape[1]
X1 = np.reshape(X, (N, ))
Y1 = np.reshape(Y, (N, ))
data = []
for k in range(0, N):
data.append([X1[k], daytype, hourtype, seat, Y1[k], discount])
print(len(data))
Z1 = np.array(kde.computeProbs(data))
Z = np.reshape(Z1, old_shape)
print(X.shape, Y.shape, Z.shape)
surf = ax.plot_surface(X,
Y,
Z,
rstride=1,
cstride=1,
cmap=cm.coolwarm,
linewidth=0,
antialiased=False)
ax.zaxis.set_major_locator(LinearLocator(10))
ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
fig.colorbar(surf, shrink=0.5, aspect=5)
ax.set_xlabel(r'#advanced days')
ax.set_ylabel(r'price kpi')
ax.set_zlabel(r'density')
plt.savefig('plots/{0}_3d.pdf'.format(pId))
plt.close(fig)
return
def visualizeTrainedModel():
'''
3d bar plot
'''
modelName = 'mix-fKDE2_cv'
modelPath = 'models/{0}'.format(modelName)
# 3000939044_mix-fKDE2_1.5.model
# mFile='models/{0}/{1}_{2}.model'.format(modelName, uId,modelName)
trainedModels = random.sample(os.listdir(modelPath), 10)
for m in trainedModels:
mFile = '{0}/{1}'.format(modelPath, m)
print(mFile)
pId = mFile.split('/')[-1].replace('.model', '').replace(
'_{0}'.format(modelName), '')
print(pId)
m = comparemethods.mixfKDEModel(pId, modelFile=mFile)
m.loadModel()
kde = m.trainedModel
# xs: advanced day
# ys: price kpi
# fixed: seat class, daytype
daytype = commonoperation.DayType.WORKDAY
hourtype = 15 / 2
seat = 1
discount = 0.8
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for c, z in zip(['r', 'g', 'b', 'y'], [0.25, 0.5, 0.75, 1]):
data = []
xs = range(0, 20, 2)
for x in xs:
data.append([x, daytype, hourtype, seat, z, discount])
ys = kde.computeProbs(data)
# You can provide either a single color or an array. To demonstrate this,
# the first bar of each set will be colored cyan.
cs = [c] * len(xs)
# cs[0] = 'c'
ax.bar(xs, ys, zs=z, zdir='y', color=cs, alpha=0.8)
# print(xs)
# print(ys)
# print(z
# break
ax.set_xlabel(r'#advanced days')
ax.set_ylabel(r'price kpi')
ax.set_zlabel(r'density')
plt.savefig('plots/{0}_3d.pdf'.format(pId))
plt.close(fig)
return
def train(self, simiDefine='Euclid'):
for u in self.userList:
if simiDefine == 'Euclid':
m6 = comparemethods.mixfKDEModel(
u,
modelName=self.modelName,
trainData=self.userTrainData[u],
testData=self.userTestData[u],
trainDataOfNeighbors=self.featureListOfNeighbors[u])
else:
m6 = comparemethods.mixfKDEModelDim(
u,
modelName=self.modelName,
trainData=self.userTrainData[u],
testData=self.userTestData[u],
trainDataOfNeighbors=self.featureListOfNeighborsDim[u])
m6.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml')
m6.train()
self.modelList[u] = m6
return 0
def modelUsers(users): # experiment 1
# users=usergeneration.generateSamplesOfActiveUsers(10000)
for u in users:
print(u)
datapreparation.prepareData(u)
traindata = datapreparation.getTraindata(u)
testdata = datapreparation.getTestdata(u)
print('only train data')
# GMM
m1 = comparemethods.GMMModel(u,
modelName='GMM1_2',
trainData=traindata,
testData=testdata)
m1.setVariables(nComponents=2)
m1.run()
m1 = comparemethods.GMMModel(u,
modelName='GMM1_3',
trainData=traindata,
testData=testdata)
m1.setVariables(nComponents=3)
m1.run()
m2 = comparemethods.fKDEModel(u,
modelName='fKDE1_Silverman',
trainData=traindata,
testData=testdata)
m2.setVariables(bandwidth='Silverman')
m2.run()
neighbors = usergeneration.getSimilarUsers(u)
featuresListOfNeighbors = []
for n in neighbors:
featuresListOfNeighbors += featureextraction.generateFeaturesList(
n)
print('with others data')
print('GMM')
m3 = comparemethods.GMMModel(u,
modelName='GMM2_2',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m3.setVariables(nComponents=2)
m3.run()
m3 = comparemethods.GMMModel(u,
modelName='GMM2_3',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m3.setVariables(nComponents=3)
m3.run()
print('fKDE2_Silverman')
m4 = comparemethods.fKDEModel(u,
modelName='fKDE2_Silverman',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m4.setVariables(bandwidth='Silverman')
m4.run()
print('mix-fKDE1_Silverman')
m5 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE1_Silverman',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m5.setVariables(bandwidth='Silverman', bandwidth1='Silverman')
m5.run()
print('mix-fKDE2_cv')
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml')
m6.run()
def modelUsers3(users):
'''
Compare mix-fKDE2_cv with different kernels
'''
for u in users:
print(u)
datapreparation.prepareData(u)
# train,test data from this user
traindata = datapreparation.getTraindata(u)
testdata = datapreparation.getTestdata(u)
neighbors = usergeneration.getSimilarUsers(u)
# feature from other users, to make mix KDE
featuresListOfNeighbors = []
for n in neighbors:
featuresListOfNeighbors += featureextraction.generateFeaturesList(
n)
m2 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_tophat',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m2.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml', kernel='tophat')
m2.run()
m3 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_epanechnikov',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m3.setVariables(bandwidth='cv_ml',
bandwidth1='cv_ml',
kernel='epanechnikov')
m3.run()
m4 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_exponential',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m4.setVariables(bandwidth='cv_ml',
bandwidth1='cv_ml',
kernel='exponential')
m4.run()
m5 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_linear',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m5.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml', kernel='linear')
m5.run()
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_cosine',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml', kernel='cosine')
m6.run()
m1 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv_gaussian',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m1.setVariables(bandwidth='cv_ml',
bandwidth1='cv_ml',
kernel='gaussian')
m1.run()
def modelUsers2(users): # experiment 2
'''
Compare different methods with different bandwidths
'''
# users=usergeneration.generateSamplesOfActiveUsers(10000)
for u in users:
print(u)
datapreparation.prepareData(u)
traindata = datapreparation.getTraindata(u)
testdata = datapreparation.getTestdata(u)
print('only train data')
# GMM
m1 = comparemethods.GMMModel(u,
modelName='GMM1_2',
trainData=traindata,
testData=testdata)
m1.setVariables(nComponents=2)
m1.run()
m1 = comparemethods.GMMModel(u,
modelName='GMM1_3',
trainData=traindata,
testData=testdata)
m1.setVariables(nComponents=3)
m1.run()
m2 = comparemethods.fKDEModel(u,
modelName='fKDE1_Silverman',
trainData=traindata,
testData=testdata)
m2.setVariables(bandwidth='Silverman')
m2.run()
m2 = comparemethods.fKDEModel(u,
modelName='fKDE1_0.5',
trainData=traindata,
testData=testdata)
m2.setVariables(bandwidth=0.5)
m2.run()
m2 = comparemethods.fKDEModel(u,
modelName='fKDE1_1',
trainData=traindata,
testData=testdata)
m2.setVariables(bandwidth=1)
m2.run()
m2 = comparemethods.fKDEModel(u,
modelName='fKDE1_1.5',
trainData=traindata,
testData=testdata)
m2.setVariables(bandwidth=1.5)
m2.run()
neighbors = usergeneration.getSimilarUsers(u)
featuresListOfNeighbors = []
for n in neighbors:
featuresListOfNeighbors += featureextraction.generateFeaturesList(
n)
# try another method: find neighbors on each dimension, each idx represents a dim
featuresListOfNeighborsDims = []
for i in range(0, len(basicmining.getProfile(u))):
neighbors = usergeneration.getSimilarUsersDim(u, i)
listTemp = []
for n in neighbors:
listTemp += (featureextraction.generateFeaturesList(n))
featuresListOfNeighborsDims.append(listTemp)
print('with others data')
m3 = comparemethods.GMMModel(u,
modelName='GMM2_2',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m3.setVariables(nComponents=2)
m3.run()
m3 = comparemethods.GMMModel(u,
modelName='GMM2_3',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m3.setVariables(nComponents=3)
m3.run()
# print('fKDE2_Silverman')
m4 = comparemethods.fKDEModel(u,
modelName='fKDE2_Silverman',
trainData=traindata +
featuresListOfNeighbors,
testData=testdata)
m4.setVariables(bandwidth='Silverman')
m4.run()
# print('mix-fKDE2_bw')
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_Silverman',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth='Silverman', bandwidth1='Silverman')
m6.run()
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_0.5',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth=0.5, bandwidth1=0.5)
m6.run()
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_1',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth=1, bandwidth1=1)
m6.run()
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_1.5',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth=1.5, bandwidth1=1.5)
m6.run()
m6 = comparemethods.mixfKDEModel(
u,
modelName='mix-fKDE2_cv',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighbors)
m6.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml')
m6.run()
# test set components according to the dim
m7 = comparemethods.mixfKDEModelDim(
u,
modelName='mix-fKDE2_cv_moreComponents',
trainData=traindata,
testData=testdata,
trainDataOfNeighbors=featuresListOfNeighborsDims)
m7.setVariables(bandwidth='cv_ml', bandwidth1='cv_ml')
m7.run()