def plotOrigVsDetrend():
data = constructData()
# Original time series
data1 = constructData()
origY = data1[1][0:len(data[1])-365]
# Detrended time series
indices = np.arange(len(data[1])-365)
detrendY = statistics.detrend(indices,data[1][0:len(data[1])-365])[0]
visualizer.comparisonPlot(2009,1,1,origY,detrendY,
"Original","Detrended",plotName="Aggregate Electric Load : Original & Detrended", yAxisName="Kilowatts")
def plotDetrended():
data = constructData()
# Plot of data after detrending with least squares regression
indices = np.arange(len(data[1]))
detrendY = statistics.detrend(indices,data[1])[0]
visualizer.yearlyPlot(detrendY,
2009,1,1,"Detrended Aggregate Electricity Demand","Residual Kilowatts")
def plotOriginal():
data = constructData()
# Plot of aggregate electricity demand over the past 5 years
section = data[1][0:len(data[1]) - 365]
visualizer.yearlyPlot(section, 2009, 1, 1,
"Average Total Electricity Load : 2009-2013",
"Kilowatts")
def plotDetrended():
data = constructData()
# Plot of data after detrending with least squares regression
indices = np.arange(len(data[1]))
detrendY = statistics.detrend(indices, data[1])[0]
visualizer.yearlyPlot(detrendY, 2009, 1, 1,
"Detrended Aggregate Electricity Demand",
"Residual Kilowatts")
def suppVectorRegress():
kernelList = ["linear","rbf",polyKernel]
names = ["linear","radial basis","poly"]
preds = []
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
#print (data[0][1430])
cutoff = len(data[0])-89 #predict march
#print cutoff
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
#print xTrain
#print xTest
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain,0.0)
statistics.estimateMissing(xTest,0.0)
# Logarithmically scale the data
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
#print indices
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
#print testIndices
detrended,slope,intercept = statistics.detrend(trainIndices,yTrain)
yTrain = detrended
for gen in range(len(kernelList)):
# Use SVR to predict test observations based upon training observations
pred = svrPredictions(xTrain,yTrain,xTest,kernelList[gen])
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices,pred,slope,intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest,trendedPred)
print "The Normalized Root-Mean Square Error is " + str(err) + " using kernel " + names[gen] + "..."
preds.append(trendedPred)
names.append("actual")
preds.append(yTest)
visualizer.comparisonPlot(2014,1,1,preds,names,plotName="Support Vector Regression Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def plotOrigVsDetrend():
data = constructData()
# Original time series
data1 = constructData()
origY = data1[1][0:len(data[1]) - 365]
# Detrended time series
indices = np.arange(len(data[1]) - 365)
detrendY = statistics.detrend(indices, data[1][0:len(data[1]) - 365])[0]
visualizer.comparisonPlot(
2009,
1,
1,
origY,
detrendY,
"Original",
"Detrended",
plotName="Aggregate Electric Load : Original & Detrended",
yAxisName="Kilowatts")
def suppVectorRegress():
kernelList = ["linear","rbf",polyKernel]
names = ["linear","radial basis","poly"]
preds = []
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
cutoff = len(data)-364
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain,0.0)
statistics.estimateMissing(xTest,0.0)
# Logarithmically scale the data
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended,slope,intercept = statistics.detrend(trainIndices,yTrain)
yTrain = detrended
for gen in range(len(kernelList)):
# Use SVR to predict test observations based upon training observations
pred = svrPredictions(xTrain,yTrain,xTest,kernelList[gen])
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices,pred,slope,intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest,trendedPred)
print "The Normalized Root-Mean Square Error is " + str(err) + " using kernel " + names[gen] + "..."
preds.append(trendedPred)
names.append("actual")
preds.append(yTest)
visualizer.comparisonPlot(2014,1,1,preds,names,plotName="Support Vector Regression Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def neuralNetwork():
# Retrieve time series data & apply preprocessing
data = constructData()
print len(data)
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
#cutoff = len(data[0])-89
cutoff = len(data[0]) - 89
#print "cutoff"+str(cutoff)
xTrain = data[0][0:cutoff]
#print xTrain[47]
print xTrain
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
#print cutoff
#print xTest[0]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain, 0.0)
statistics.estimateMissing(xTest, 0.0)
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended, slope, intercept = statistics.detrend(trainIndices, yTrain)
yTrain = detrended
dimensions = [6, 10, 12]
neurons = [30, 50, 50]
names = []
for x in range(len(dimensions)):
s = "d=" + str(dimensions[x]) + ",h=" + str(neurons[x])
names.append(s)
preds = []
for x in range(len(dimensions)):
# Perform dimensionality reduction on the feature vectors
pca = PCA(n_components=dimensions[x])
pca.fit(xTrain)
xTrainRed = pca.transform(xTrain)
xTestRed = pca.transform(xTest)
pred = fit_predict(xTrainRed, yTrain, xTestRed, 40, neurons[x])
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices, pred, slope,
intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest, trendedPred)
# Append computed predictions to list for classifier predictions
preds.append(trendedPred)
print "The NRMSE for the neural network is " + str(err) + "..."
preds.append(yTest)
names.append("actual")
visualizer.comparisonPlot(
2014,
1,
1,
preds,
names,
plotName="Neural Network Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def plotPeriodogram():
data = constructData()
visualizer.periodogramPlot(
data[1][len(data[1]) - 730:len(data[1]) - 365],
"Periodogram of Average Total Electricity Load : 2013")
def plotLag():
data = constructData()
visualizer.lagPlot(data[1][0:len(data[1]) - 365],
"Average Total Electricity Load Lag : 2009-2013")
def plotCorrelogram():
data = constructData()
visualizer.autoCorrPlot(
data[1][len(data[1]) - 730:len(data[1]) - 365],
"Average Total Electricity Load Autocorrelations : 2013")
def plotOriginal():
data = constructData()
# Plot of aggregate electricity demand over the past 5 years
section = data[1][0:len(data[1])-365]
visualizer.yearlyPlot(section,
2009,1,1,"Average Total Electricity Load : 2009-2013","Kilowatts")
def plotCorrelogram():
data = constructData()
visualizer.autoCorrPlot(data[1][len(data[1])-730:len(data[1])-365],"Average Total Electricity Load Autocorrelations : 2013")
def clustering():
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
cutoff = len(data)-364
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain,0.0)
statistics.estimateMissing(xTest,0.0)
# Logarithmically scale the data
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended,slope,intercept = statistics.detrend(trainIndices,yTrain)
yTrain = detrended
# Compute centroids and labels of data
cward_7,lward_7 = hierarchicalClustering(xTrain,7)
cward_365,lward_365 = hierarchicalClustering(xTrain,365)
ckmeans_7,lkmeans_7 = kMeansClustering(xTrain,7)
ckmeans_365,lkmeans_365 = kMeansClustering(xTrain,365)
c = [cward_7,cward_365,ckmeans_7,ckmeans_365]
l = [lward_7,lward_365,lkmeans_7,lkmeans_365]
algNames = ["agglomerative(7)","agglomerative(365)","k-means(7)","k-means(365)"]
preds = []
for t in range(len(c)):
# The centroids computed by the current clustering algorithm
centroids = c[t]
# The labels for the examples defined by the current clustering assignment
labels = l[t]
# Separate the training samples into cluster sets
clusterSets = []
# Time labels for the examples, separated into clusters
timeLabels = []
for x in range(len(centroids)):
clusterSets.append([])
for x in range(len(labels)):
# Place the example into its cluster
clusterSets[labels[x]].append((xTrain[x],yTrain[x]))
# Compute predictions for each of the test examples
pred = predictClustering(centroids,clusterSets,xTest,"euclidean")
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices,pred,slope,intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest,trendedPred)
# Add to list of predictions
preds.append(trendedPred)
print "The Normalized Root-Mean Square Error is " + str(err) + " using algorithm " + algNames[t] + "..."
algNames.append("actual")
preds.append(yTest)
visualizer.comparisonPlot(2014,1,1,preds,algNames,
plotName="Clustering Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def clustering():
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
cutoff = len(data) - 364
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain, 0.0)
statistics.estimateMissing(xTest, 0.0)
# Logarithmically scale the data
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended, slope, intercept = statistics.detrend(trainIndices, yTrain)
yTrain = detrended
# Compute centroids and labels of data
cward_7, lward_7 = hierarchicalClustering(xTrain, 7)
cward_365, lward_365 = hierarchicalClustering(xTrain, 365)
ckmeans_7, lkmeans_7 = kMeansClustering(xTrain, 7)
ckmeans_365, lkmeans_365 = kMeansClustering(xTrain, 365)
c = [cward_7, cward_365, ckmeans_7, ckmeans_365]
l = [lward_7, lward_365, lkmeans_7, lkmeans_365]
algNames = [
"agglomerative(7)", "agglomerative(365)", "k-means(7)", "k-means(365)"
]
preds = []
for t in range(len(c)):
# The centroids computed by the current clustering algorithm
centroids = c[t]
# The labels for the examples defined by the current clustering assignment
labels = l[t]
# Separate the training samples into cluster sets
clusterSets = []
# Time labels for the examples, separated into clusters
timeLabels = []
for x in range(len(centroids)):
clusterSets.append([])
for x in range(len(labels)):
# Place the example into its cluster
clusterSets[labels[x]].append((xTrain[x], yTrain[x]))
# Compute predictions for each of the test examples
pred = predictClustering(centroids, clusterSets, xTest, "euclidean")
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices, pred, slope,
intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest, trendedPred)
# Add to list of predictions
preds.append(trendedPred)
print "The Normalized Root-Mean Square Error is " + str(
err) + " using algorithm " + algNames[t] + "..."
algNames.append("actual")
preds.append(yTest)
visualizer.comparisonPlot(
2014,
1,
1,
preds,
algNames,
plotName="Clustering Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def plotPeriodogram():
data = constructData()
visualizer.periodogramPlot(data[1][len(data[1])-730:len(data[1])-365],
"Periodogram of Average Total Electricity Load : 2013")
def plotLag():
data = constructData()
visualizer.lagPlot(data[1][0:len(data[1])-365],"Average Total Electricity Load Lag : 2009-2013")
def gaussianProcesses():
corrMods = [
'cubic', 'squared_exponential', 'absolute_exponential', 'linear'
]
preds = []
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
cutoff = len(data) - 364
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain, 0.0)
statistics.estimateMissing(xTest, 0.0)
# Logarithmically scale the data
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended, slope, intercept = statistics.detrend(trainIndices, yTrain)
yTrain = detrended
for gen in range(len(corrMods)):
# Use GPR to predict test observations based upon training observations
pred = gaussProcPred(xTrain, yTrain, xTest, corrMods[gen])
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices, pred, slope,
intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest, trendedPred)
print "The Normalized Root-Mean Square Error is " + str(
err) + " using covariance function " + corrMods[gen] + "..."
preds.append(trendedPred)
corrMods.append("actual")
data = constructData()
cutoff = len(data) - 364
yTest = data[1][cutoff:]
preds.append(yTest)
visualizer.comparisonPlot(
2014,
1,
1,
preds,
corrMods,
plotName="Gaussian Process Regression Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")
def neuralNetwork():
# Retrieve time series data & apply preprocessing
data = constructData()
# 2014 had 365 days, but we take the last 364 days since
# the last day has no numerical value
cutoff = len(data)-364
xTrain = data[0][0:cutoff]
yTrain = data[1][0:cutoff]
xTest = data[0][cutoff:]
yTest = data[1][cutoff:]
# Fill in missing values denoted by zeroes as an average of
# both neighbors
statistics.estimateMissing(xTrain,0.0)
statistics.estimateMissing(xTest,0.0)
xTrain = [[math.log(y) for y in x] for x in xTrain]
xTest = [[math.log(y) for y in x] for x in xTest]
yTrain = [math.log(x) for x in yTrain]
# Detrend the time series
indices = np.arange(len(data[1]))
trainIndices = indices[0:cutoff]
testIndices = indices[cutoff:]
detrended,slope,intercept = statistics.detrend(trainIndices,yTrain)
yTrain = detrended
dimensions = [6,10,12]
neurons = [30,50,50]
names = []
for x in range(len(dimensions)):
s = "d=" + str(dimensions[x]) + ",h=" + str(neurons[x])
names.append(s)
preds = []
for x in range(len(dimensions)):
# Perform dimensionality reduction on the feature vectors
pca = PCA(n_components=dimensions[x])
pca.fit(xTrain)
xTrainRed = pca.transform(xTrain)
xTestRed = pca.transform(xTest)
pred = fit_predict(xTrainRed,yTrain,xTestRed,40,neurons[x])
# Add the trend back into the predictions
trendedPred = statistics.reapplyTrend(testIndices,pred,slope,intercept)
# Reverse the normalization
trendedPred = [math.exp(x) for x in trendedPred]
# Compute the NRMSE
err = statistics.normRmse(yTest,trendedPred)
# Append computed predictions to list for classifier predictions
preds.append(trendedPred)
print "The NRMSE for the neural network is " + str(err) + "..."
preds.append(yTest)
names.append("actual")
visualizer.comparisonPlot(2014,1,1,preds,names,plotName="Neural Network Load Predictions vs. Actual",
yAxisName="Predicted Kilowatts")