def tuneTsneBisecting(createOutput):
numTrials = 5
X = common.readDigitsFile()
print("Shape of Digits file = " + str(X.shape))
print("Starting t-SNE.")
#pca = PCA(n_components=50)
#X_temp = pca.fit_transform(X)
tsne = TSNE(n_components=2,
init='random',
random_state=0,
perplexity=35,
learning_rate=400,
n_iter=2000)
X_new = tsne.fit_transform(X)
#X_new = digitutil.preprocessDownsampling(X)
print("Finished t-SNE, and got X_new. Shape = " + str(X_new.shape))
#print("First 10 rows are: ")
#print(X_new[0:10, :])
#numTrials = 10
#X_normalized = normalize(X, norm='l2', axis=0)
distanceFunc = common.euclideanDistanceFunction
centerFunc = common.digitsDataCenterFunction
clusterLabels = np.array(list(range(1, 11)), dtype=np.int8)
bestAssignments = None
bestErrorTotal = None
for z in range(numTrials):
model = kmclustering.BisectingKMeansClusteringModel(
distanceFunc, centerFunc, 3)
model.runBisectingKMeansClustering(X_new, clusterLabels,
kminit.initKMeansSampling)
#model = kmclustering.BasicKMeansClusteringModel(distanceFunc, centerFunc, 100)
#initFunc = lambda q, z: kminit.initKMeansSampling(q, z)
#assignments, centers = kminit.initKMeansSampling(X_normalized, clusterLabels, distanceFunc)
#model.runBasicKMeansClustering(X_new, clusterLabels, kminit.initKMeansSampling)
print("====== Done with Trial # " + str(z + 1) + " / " +
str(numTrials) + ", Error Total = " +
str(model.finalClusterErrorTotal) + " ======")
#print("Assignments = " + str(model.finalClusterAssignments))
#print("Centers = " + str(model.finalClusterCenters))
#print("Error Map = " + str(model.finalClusterErrorMap))
#print()
if bestErrorTotal is None or model.finalClusterErrorTotal < bestErrorTotal:
bestErrorTotal = model.finalClusterErrorTotal
bestAssignments = model.finalClusterAssignments
#print("Improved total SSE! New Value = " + str(bestErrorTotal))
print("Finished K-Means with t-SNE, P=. Best Error Total = " +
str(bestErrorTotal))
#print("Best Assignments = " + str(bestAssignments))
if createOutput:
common.writeResultsFile(bestAssignments)
def chartClusterErrorVsClusterCount():
X = common.readDigitsFile()
X = X / 255.0
print("Shape of Digits file = " + str(X.shape))
print("Starting t-SNE.")
tsne = TSNE(n_components=2,
init='random',
random_state=0,
perplexity=100,
learning_rate=400,
n_iter=2000)
X_new = tsne.fit_transform(X)
print("Finished t-SNE, and got X_new. Shape = " + str(X_new.shape))
distanceFunc = common.euclideanDistanceFunction
centerFunc = common.digitsDataCenterFunction
numTrials = 5
errorMapForClusterSize = {}
clusterSizeList = list(range(2, 21, 2))
for clusterSize in clusterSizeList:
errorSum = 0
clusterLabels = np.array(list(range(1, clusterSize + 1)),
dtype=np.int8)
for z in range(numTrials):
model = kmclustering.BasicKMeansClusteringModel(
distanceFunc, centerFunc, 100)
model.runBasicKMeansClustering(X_new, clusterLabels,
kminit.initKMeansSampling)
print("====== Done with Trial # " + str(z + 1) + " / " + str(numTrials) + " for K = " + str(clusterSize) + \
", Error Total = " + str(model.finalClusterErrorTotal) + " ======")
errorSum += model.finalClusterErrorTotal
avgError = errorSum / numTrials
errorMapForClusterSize[clusterSize] = avgError
print("Done with K-Means.")
for clusterSize in clusterSizeList:
print("For K = " + str(clusterSize) + " Error is: " +
str(errorMapForClusterSize[clusterSize]))
def submission02(createOutput, numTrials):
X = common.readDigitsFile()
print("Shape of Digits file = " + str(X.shape))
pca = PCA(n_components=50)
X_temp = pca.fit_transform(X)
tsne = TSNE(n_components=2, init='pca')
X_new = tsne.fit_transform(X_temp)
#X_new = digitutil.preprocessDownsampling(X)
print("Got X_new. Shape = " + str(X_new.shape))
#print("First 10 rows are: ")
#print(X_new[0:10, :])
#numTrials = 10
X_normalized = normalize(X, norm='l2', axis=0)
distanceFunc = common.euclideanDistanceFunction
centerFunc = common.digitsDataCenterFunction
clusterLabels = np.array(list(range(1, 11)), dtype=np.int8)
bestAssignments = None
bestErrorTotal = None
for z in range(numTrials):
model = kmclustering.BasicKMeansClusteringModel(
distanceFunc, centerFunc, 55)
#initFunc = lambda q, z: kminit.initKMeansSampling(q, z)
#assignments, centers = kminit.initKMeansSampling(X_normalized, clusterLabels, distanceFunc)
model.runBasicKMeansClustering(X_normalized, clusterLabels,
kminit.initKMeansSampling)
print("====== Done with Trial # " + str(z + 1) + " / " +
str(numTrials) + ", Error Total = " +
str(model.finalClusterErrorTotal) + " ======")
#print("Assignments = " + str(model.finalClusterAssignments))
#print("Centers = " + str(model.finalClusterCenters))
#print("Error Map = " + str(model.finalClusterErrorMap))
#print()
if bestErrorTotal is None or model.finalClusterErrorTotal < bestErrorTotal:
bestErrorTotal = model.finalClusterErrorTotal
bestAssignments = model.finalClusterAssignments
#print("Improved total SSE! New Value = " + str(bestErrorTotal))
print("Finished K-Means. Best Error Total = " + str(bestErrorTotal))
#print("Best Assignments = " + str(bestAssignments))
if createOutput:
common.writeResultsFile(bestAssignments)