def main():
data_file = "data/winequality-red.csv"
x, y = extractData(data_file)
decompose = Decomposing()
decompose.pca_dim_reduction(x, y, 'wine', 11)
decompose.pca_eval(x, y, 'wine', 5, 'Rating')
decompose.ica_dim_reduction(x, y, 'wine', 11)
decompose.ica_eval(x, y, 'wine', 11, 'Rating')
n = decompose.rp_dim_reduction(x, y, 'wine', 11)
print('Optimal n for RP(Wine data set) : {} '.format(str(n)))
decompose.rp_eval(x, y, 'wine', n, 'Rating')
k = decompose.sk_dim_reduction(x, y, 'wine', 11)
# decompose.sk_eval(x, y, 'wine', k)
data_file = "data/default_of_credit_card_clients.csv"
x, y = extractData(data_file)
decompose = Decomposing()
decompose.pca_dim_reduction(x, y, 'default', 24)
decompose.pca_eval(x, y, 'default', 10, 'Defaulted')
decompose.ica_dim_reduction(x, y, 'default', 24)
decompose.ica_eval(x, y, 'default', 6, 'Defaulted')
n = decompose.rp_dim_reduction(x, y, 'default', 24)
print('Optimal n for RP(CC Default data set) : {} '.format(str(n)))
decompose.rp_eval(x, y, 'default', n, 'Defaulted')
k = decompose.sk_dim_reduction(x, y, 'default', 24)
def classify(self, data_file, encode, label):
X, Y = extractData(data_file)
enc = LabelEncoder()
if encode:
rows, cols = X.shape
for c in range(cols):
if not str(X.iloc[1][c]).isnumeric():
enc.fit(X.iloc[:, c])
X.iloc[:, c] = enc.transform(X.iloc[:, c])
enc.fit(Y)
Y = enc.transform(Y)
train_x, test_x, train_y, test_y = train_test_split(X, Y, test_size=0.3)
nn = MLPClassifier()
parameter_grid = {'activation' : ['identity', 'logistic', 'tanh', 'relu'],
'solver' : ['lbfgs', 'sgd', 'adam'],
'hidden_layer_sizes': [
(5,),(10,),(15,),(20,)
]
}
start = timer()
classifier, grid_search = getBestModel(nn, parameter_grid, train_x, train_y)
classify_model = classifier.fit(train_x, train_y)
pred_y = classify_model.predict(test_x)
end = timer()
print('Elapsed time of train and test : ' + str(end - start))
accuracy = metrics.accuracy_score(test_y, pred_y) * 100
print('Accuracy of Neural Network = {:.2f}%'.format(accuracy))
plotValidationCurve("Neural Network", label, grid_search, train_x, train_y, parameter_grid)
plotLearningCurve("Neural Network", label, classifier, X, Y)
plotPerformance(test_y, pred_y, label, 'Algorithm: Neural Network')
def main():
results = pd.DataFrame(columns=[
'Data Set', 'Cluster Algo.', '# Clusters', 'Mutual Info Score'
])
cluster = Clustering()
data_file = "data/winequality-red.csv"
x, y = extractData(data_file)
x = pd.DataFrame(preprocessing.scale(x), columns=x.columns)
cluster.kMeansCluster('wine', x, y, 6, 10)
km = cluster.kmeansFitBestModel(6, 10, 'wine', x, 5)
score = adjusted_mutual_info_score(km.labels_, y)
results = append_results(results, ['Wine Quality', 'k-Means', 5, score])
cluster.emCluster('wine', x, y, 6, 10)
em = cluster.emFitBestModel(6, 10, 'wine', 5, x)
score = adjusted_mutual_info_score(em.predict(x), y)
results = append_results(results,
['Wine Quality', 'Exp. Maximization', 5, score])
data_file = "data/default_of_credit_card_clients.csv"
x, y = extractData(data_file)
x = pd.DataFrame(preprocessing.scale(x), columns=x.columns)
bill_cols = [
'BILL_AMT1', 'BILL_AMT2', 'BILL_AMT3', 'BILL_AMT4', 'BILL_AMT5',
'BILL_AMT6'
]
pmt_cols = [
'PAY_AMT1', 'PAY_AMT2', 'PAY_AMT3', 'PAY_AMT4', 'PAY_AMT5', 'PAY_AMT6'
]
num_cols = x.shape[1]
x['total_bill'] = x.loc[:, bill_cols].sum(axis=1)
x['total_pmt'] = x.loc[:, pmt_cols].sum(axis=1)
cluster.kMeansCluster('default', x, y, num_cols, num_cols + 1)
km = cluster.kmeansFitBestModel(num_cols, num_cols + 1, 'default', x, 10)
score = adjusted_mutual_info_score(km.labels_, y)
results = append_results(results, ['CC Default', 'k-Means', 10, score])
cluster.emCluster('default', x, y, num_cols, num_cols + 1)
em = cluster.emFitBestModel(num_cols, num_cols + 1, 'default', 15, x)
score = adjusted_mutual_info_score(em.predict(x), y)
results = append_results(results,
['CC Default', 'Exp. Maximization', 15, score])
print(results)
print(results.to_latex())
def classify(self, data_file, encode, label):
# data = pd.read_csv(data_file)
X, Y = extractData(data_file)
enc = LabelEncoder()
if encode:
rows, cols = X.shape
for c in range(cols):
if not str(X.iloc[1][c]).isnumeric():
enc.fit(X.iloc[:, c])
X.iloc[:, c] = enc.transform(X.iloc[:, c])
enc.fit(Y)
Y = enc.transform(Y)
#
train_x, test_x, train_y, test_y = train_test_split(X,
Y,
test_size=0.2,
random_state=123)
#Find best model
dct = tree.DecisionTreeClassifier()
parameter_grid = {
'criterion': ['gini', 'entropy'],
'splitter': ['best', 'random'],
'max_depth': range(1, 10),
'max_features': range(1, 5),
'min_samples_leaf': range(1, 5)
}
start = timer()
classifier, grid_search = getBestModel(dct, parameter_grid, train_x,
train_y)
classify_model = classifier.fit(train_x, train_y)
pred_y = classify_model.predict(test_x)
end = timer()
print('Elapsed time of train and test : ' + str(end - start))
accuracy = grid_search.best_score_ * 100
print('Accuracy of Decision Tree(depth={}) = {:.2f}%'.format(
classify_model.get_depth(), accuracy))
plotValidationCurve("Decision Tree", label, grid_search, train_x,
train_y, parameter_grid)
plotLearningCurve("Decision Tree", label, classifier, X, Y)
plotPerformance(test_y, pred_y, label, 'Algorithm: Decision Tree')
def classifyWithBoost(self, data_file, encode, label):
X, Y = extractData(data_file)
enc = LabelEncoder()
if encode:
rows, cols = X.shape
for c in range(cols):
if not str(X.iloc[1][c]).isnumeric():
enc.fit(X.iloc[:, c])
X.iloc[:, c] = enc.transform(X.iloc[:, c])
enc.fit(Y)
Y = enc.transform(Y)
train_x, test_x, train_y, test_y = train_test_split(X,
Y,
test_size=0.3,
random_state=123)
dctb = GradientBoostingClassifier(random_state=123)
parameter_grid = {
'learning_rate': [0.2, 0.3, 0.5],
'max_depth': [2, 3, 4, 5]
}
start = timer()
classifier, grid_search = getBestModel(dctb, parameter_grid, train_x,
train_y)
plotValidationCurve("Decision Tree with Boost", label, grid_search,
train_x, train_y, parameter_grid)
plotLearningCurve("Decision Tree with Boost", label, classifier, X, Y)
classify_model = classifier.fit(X, Y)
pred_y = classify_model.predict(test_x)
end = timer()
print('Elapsed time of train and test : ' + str(end - start))
accuracy = classify_model.score(test_x, test_y) * 100
print('Accuracy of GradientBoostingClassifier = {:.2f}%'.format(
accuracy))
plotPerformance(test_y, pred_y, label,
'Algorithm: Decision Tree with Boost')
def classify(self, data_file, encode, label):
X, Y = extractData(data_file)
enc = LabelEncoder()
if encode:
rows, cols = X.shape
for c in range(cols):
if not str(X.iloc[1][c]).isnumeric():
enc.fit(X.iloc[:, c])
X.iloc[:, c] = enc.transform(X.iloc[:, c])
enc.fit(Y)
Y = enc.transform(Y)
#
train_x, test_x, train_y, test_y = train_test_split(X,
Y,
test_size=0.3,
random_state=123)
knn = KNeighborsClassifier()
parameter_grid = {'n_neighbors': range(1, 10)}
start = timer()
classifier, grid_search = getBestModel(knn, parameter_grid, train_x,
train_y)
classify_model = classifier.fit(train_x, train_y)
pred_y = classify_model.predict(test_x)
accuracy = classify_model.score(test_x, test_y) * 100
end = timer()
print('Elapsed time of train and test : ' + str(end - start))
print('Accuracy of KNN = {:.2f}%'.format(accuracy))
plotValidationCurve("KNeighbours", label, grid_search, train_x,
train_y, parameter_grid)
plotLearningCurve("KNeighbours", label, classifier, X, Y)
plotPerformance(test_y, pred_y, label,
'Algorithm: k-Nearest Neighbors')
def data_prep(self, test_size=0.3):
x, y = extractData("data/default_of_credit_card_clients.csv")
self.num_classes = len(np.unique(y))
self.train_x, self.test_x, self.train_y, self.test_y = train_test_split(
x, y, test_size=test_size)
def main():
exp_results = pd.DataFrame(columns=['Data Set','Dim. Red. Algo','Cluster Algo.','# Clusters','Mutual Info Score'])
data_file = "data/winequality-red.csv"
x, y = extractData(data_file)
x = pd.DataFrame(preprocessing.scale(x), columns=x.columns)
dimreduce = Decomposing()
cluster = Clustering()
pca_result = pd.DataFrame(dimreduce.pca_eval(x,y,'experiment/wine/pca',5,'Rating')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/wine/pca', pca_result.copy(), 5)
score = adjusted_mutual_info_score(km.labels_,y)
print('Wine : PCA - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality','PCA','k-Means',5,score])
em = cluster.emFitBestModel(1,2,'experiment/wine/pca',5,pca_result)
score = adjusted_mutual_info_score(em.predict(pca_result), y)
print('Wine : PCA - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'PCA', 'Exp. Maximization', 5, score])
ica_result = pd.DataFrame(dimreduce.ica_eval(x, y, 'experiment/wine/ica', 11, 'Rating')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/wine/ica', ica_result.copy(), 5)
score = adjusted_mutual_info_score(km.labels_, y)
print('Wine : ICA - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'ICA', 'k-Means', 5, score])
em = cluster.emFitBestModel(1, 2, 'experiment/wine/ica', 5, ica_result)
score = adjusted_mutual_info_score(em.predict(ica_result), y)
print('Wine : ICA - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'ICA', 'Exp. Maximization', 5, score])
rp_result = pd.DataFrame(dimreduce.rp_eval(x, y, 'experiment/wine/rp', 11, 'Rating')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/wine/rp', rp_result.copy(), 5)
score = adjusted_mutual_info_score(km.labels_, y)
print('Wine : RP - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'RP', 'k-Means', 5, score])
em = cluster.emFitBestModel(1, 2, 'experiment/wine/rp', 5, rp_result)
score = adjusted_mutual_info_score(em.predict(rp_result), y)
print('Wine : RP - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'RP', 'Exp. Maximization', 5, score])
sk_result = pd.DataFrame(dimreduce.sk_eval(x, y, 8)[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/wine/sk', sk_result.copy(), 4)
score = adjusted_mutual_info_score(km.labels_, y)
print('Wine : SK - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'Select-K', 'k-Means', 4, score])
em = cluster.emFitBestModel(1, 2, 'experiment/wine/sk', 5, sk_result)
score = adjusted_mutual_info_score(em.predict(sk_result), y)
print('Wine : SK - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['Wine Quality', 'Select-K', 'Exp. Maximization', 5, score])
data_file = "data/default_of_credit_card_clients.csv"
x, y = extractData(data_file)
x = pd.DataFrame(preprocessing.scale(x), columns=x.columns)
pca_result = pd.DataFrame(dimreduce.pca_eval(x, y, 'experiment/default/pca', 10,'Defaulted')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/default/pca', pca_result, 5)
score = adjusted_mutual_info_score(km.labels_, y)
print('Default : PCA - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'PCA', 'k-Means', 5, score])
em = cluster.emFitBestModel(1, 2, 'experiment/default/pca', 5, pca_result)
score = adjusted_mutual_info_score(em.predict(pca_result), y)
print('Default : PCA - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'PCA', 'Exp. Maximization', 5, score])
ica_result = pd.DataFrame(dimreduce.ica_eval(x, y, 'experiment/default/ica', 6, 'Defaulted')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/default/ica', ica_result, 4)
score = adjusted_mutual_info_score(km.labels_, y)
print('Default : ICA - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'ICA', 'k-Means', 6, score])
em = cluster.emFitBestModel(1, 2, 'experiment/default/ica', 4, ica_result)
score = adjusted_mutual_info_score(em.predict(ica_result), y)
print('Default : ICA - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'ICA', 'Exp. Maximization', 4, score])
rp_result = pd.DataFrame(dimreduce.rp_eval(x, y, 'experiment/default/rp', 24, 'Defaulted')[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/default/rp', rp_result, 10)
score = adjusted_mutual_info_score(km.labels_, y)
print('Default : RP - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'RP', 'k-Means', 10, score])
em = cluster.emFitBestModel(1, 2, 'experiment/default/rp', 10, rp_result)
score = adjusted_mutual_info_score(em.predict(rp_result), y)
print('Default : RP - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'RP', 'Exp. Maximization', 10, score])
sk_result = pd.DataFrame(dimreduce.sk_eval(x, y, 6)[0])
km = cluster.kmeansFitBestModel(1, 2, 'experiment/default/sk', sk_result, 4)
score = adjusted_mutual_info_score(km.labels_, y)
print('Default : SK - Kmeans : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'Select-K', 'k-Means', 4, score])
em = cluster.emFitBestModel(1, 2, 'experiment/default/sk', 4, rp_result)
score = adjusted_mutual_info_score(em.predict(rp_result), y)
print('Default : SK - EM : Score = {}'.format(score))
exp_results = append_results(exp_results, ['CC Default', 'Select-K', 'Exp. Maximization', 4, score])
print(exp_results)
print(exp_results.to_latex())
##### NN
train_x, test_x, train_y, test_y = train_test_split(x, y, test_size=0.3)
accuracy_raw = nn_classifier(train_x, train_y, test_x, test_y)
pca_result, pca = dimreduce.pca_eval(train_x, train_y, 'nn', 10, 'Defaulted')
test_data = pca.transform(test_x)
accuracy_pca = nn_classifier(pca_result, train_y, test_data, test_y)
ica_result, ica = dimreduce.ica_eval(train_x, train_y, 'nn', 6, 'Defaulted')
test_data = ica.transform(test_x)
accuracy_ica = nn_classifier(ica_result, train_y, test_data, test_y)
rp_result, rp = dimreduce.rp_eval(train_x, train_y, 'nn', 24, 'Defaulted')
test_data = rp.transform(test_x)
accuracy_rp = nn_classifier(rp_result, train_y, test_data, test_y)
sk_result, sk = dimreduce.sk_eval(train_x, train_y, 4)
test_data = sk.transform(test_x)
accuracy_sk = nn_classifier(sk_result, train_y, test_data, test_y)
plot_results("Dimension Reduction Algo. Accuracy",'dim_red_accuracy',
['Original','PCA','ICA','RP','SK'],
[accuracy_raw,accuracy_pca,accuracy_ica,accuracy_rp,accuracy_sk],'Dim. Red. Algorithms','Accuracy')
km = cluster.kmeansFitBestModel(1,2,'nn',train_x,10)
km_data = km.fit_transform(train_x)
test_data = km.transform(test_x)
accuracy_km = nn_classifier(km_data,train_y,test_data, test_y)
em = cluster.emFitBestModel(1, 2, 'nn', 15,train_x)
em_train_labels = em.predict(train_x)
em_train_ohc = one_hot_encode(em_train_labels, 15)
em_train = np.concatenate((train_x, em_train_ohc), 1)
# one hot encode cluster labels to val set
em_test_labels = em.predict(test_x)
em_test_ohc = one_hot_encode(em_test_labels, 15)
em_test = np.concatenate((test_x, em_test_ohc), 1)
# scale data
scaler = preprocessing.StandardScaler().fit(em_train)
em_data = scaler.transform(em_train)
test_data = scaler.transform(em_test)
accuracy_em = nn_classifier(em_data, train_y, test_data, test_y)
plot_results("Clustering Accuracy", 'cluster_accuracy',['Original', 'k-Means', 'Ex. Maximization'],
[accuracy_raw, accuracy_km, accuracy_em], 'Clustering Algorithms',
'Accuracy')