def exercise2():
breast_cancer = pd.read_csv('breast_cancer.csv')
X, y = split_dataset_transformed(breast_cancer, 'Class', ['?'])
X_train, X_test, y_train, y_test = split_train_test(X, y)
# 2.a
clf = RandomForestClassifier()
clf_stats = classifier_statistics(clf, X_train, X_test, y_train, y_test)
print_dict(clf_stats, ['predicted'])
# 2.b
numb_trees = np.arange(10, 201, step=10)
for trees in numb_trees:
print('Experimenting with {} number of trees'.format(trees))
clf = RandomForestClassifier(n_estimators=trees)
clf_stats = classifier_statistics(clf, X_train, X_test, y_train,
y_test)
print_dict(clf_stats, ['predicted'])
# 2.c
depths = np.arange(5, 20)
for dep in depths:
print('Experimenting with {} depth'.format(dep))
clf = RandomForestClassifier(max_depth=dep)
clf_stats = classifier_statistics(clf, X_train, X_test, y_train,
y_test)
print_dict(clf_stats, ['predicted'])
def exercise3():
credit = pd.read_csv('credit.csv')
X, y = split_dataset_transformed(credit, 'class')
X_train, X_test, y_train, y_test = split_train_test(X, y)
min_samples = np.arange(2, 11)
for samples in min_samples:
print('Experimenting with {} number of instances to split'.format(
samples))
print('Train data')
clf = DecisionTreeClassifier(min_samples_split=samples)
clf_stats = classifier_statistics(clf, X_train, X_train, y_train,
y_train)
print_dict(clf_stats, ['predicted'])
print('Test data')
clf = DecisionTreeClassifier(min_samples_split=samples)
clf_stats = classifier_statistics(clf, X_train, X_test, y_train,
y_test)
print_dict(clf_stats, ['predicted'])
print()
print()
numb_trees = np.arange(10, 201, step=10)
for trees in numb_trees:
print('Experimenting with {} number of trees'.format(trees))
print('Train data')
clf = RandomForestClassifier(n_estimators=trees)
clf_stats = classifier_statistics(clf, X_train, X_train, y_train,
y_train)
print_dict(clf_stats, ['predicted'])
print('Test data')
clf = RandomForestClassifier(n_estimators=trees)
clf_stats = classifier_statistics(clf, X_train, X_test, y_train,
y_test)
print_dict(clf_stats, ['predicted'])
print()
print()
def naive_bayes(): gnb = GaussianNB() for d in data: X, y = split_dataset_transformed(d[0], 'consensus') X_train, X_test, y_train, y_test = split_train_test(X, y) res = classifier_statistics(gnb, X_train, X_test, y_train, y_test) export_file(res, d[1], 'Naive bayes', "")
def compare_baseline_clf():
neighbors = [3, 5, 10]
estimators = [25, 50, 100]
naive_bayes = [(GaussianNB(), 'Gaussian'),
(MultinomialNB(), 'Multinomial'),
(BernoulliNB(), 'Bernoulli')]
knns = [(KNeighborsClassifier(n_neighbors=x),
'K Nearest Neighbors {}'.format(x)) for x in neighbors]
random_forests = [(RandomForestClassifier(n_estimators=x),
'Random Forest {}'.format(x)) for x in estimators]
classifiers = {
'Naive Bayes': naive_bayes,
'K Nearest Neighbors': knns,
'Random Forest': random_forests
}
CLASSIFIER = 'Classifier'
measures_dict = {}
i = 0
for model_type in classifiers:
for specific in classifiers[model_type]:
clf, parameter = specific
res = classifier_statistics(clf, X_train, X_test, y_train, y_test)
conf_matrix = res['confusion_matrix']
score(conf_matrix)
accuracy = res['accuracy']
sensibility = res['sensibility']
specificity = res['specificity']
measures_dict[i] = {
CLASSIFIER: parameter,
'Measure': 'Accuracy',
'Value': accuracy
}
i += 1
measures_dict[i] = {
CLASSIFIER: parameter,
'Measure': 'Sensibility',
'Value': sensibility
}
i += 1
measures_dict[i] = {
CLASSIFIER: parameter,
'Measure': 'Specificity',
'Value': specificity
}
i += 1
measures = pd.DataFrame.from_dict(measures_dict, "index")
measures.to_csv('plot_data/initial_results.csv')
plt.figure(figsize=(22, 6))
ax = sns.barplot(x=CLASSIFIER, y='Value', hue='Measure', data=measures)
#plt.savefig('images/initial_results.pdf')
plt.clf()
def knn (): n_neighbors_values = [2,3,10] for d in data: for n in n_neighbors_values: X, y = split_dataset_transformed(d[0], 'consensus') X_train, X_test, y_train, y_test = split_train_test(X, y) neigh = KNeighborsClassifier(n_neighbors=n) res = classifier_statistics(neigh, X_train, X_test, y_train, y_test) export_file(res, d[1], 'KNN', "k=" + str(n))
cost1 = 10
cost2 = 500
conf_matrix = results['confusion_matrix']
fp = conf_matrix[0][1]
fn = conf_matrix[1][0]
total_cost = cost1*fp + cost2*fn
print('Total cost aachived: {}'.format(total_cost))
"""
"""
probably something is wrong here, the results seem to be too good
"""
clf = KNeighborsClassifier(n_neighbors=10)
results = classifier_statistics(clf, X_train_res, X_test, y_train_res, y_test)
print_dict(results, excluded_keys=['predicted'])
cost1 = 10
cost2 = 500
conf_matrix = results['confusion_matrix']
fp = conf_matrix[0][1]
fn = conf_matrix[1][0]
total_cost = cost1*fp + cost2*fn
print('Total cost aachived: {}'.format(total_cost))
def naive_bayes_balenced(): gnb = GaussianNB() for d in data: X_train, X_test, y_train, y_test, X_train_res, y_train_res = balance_dataset(d[0], 'consensus') res = classifier_statistics(gnb, X_train_res, X_test, y_train_res.ravel(), y_test.ravel()) export_file(res, d[1], 'Naive bayes balenced', "")
def knn (): n_neighbors_values = [2,3,10] for d in data: for n in n_neighbors_values: X, y = split_dataset_transformed(d[0], 'consensus') X_train, X_test, y_train, y_test = split_train_test(X, y) neigh = KNeighborsClassifier(n_neighbors=n) res = classifier_statistics(neigh, X_train, X_test, y_train, y_test) export_file(res, d[1], 'KNN', "k=" + str(n)) def naive_bayes(): gnb = GaussianNB() for d in data: X, y = split_dataset_transformed(d[0], 'consensus') X_train, X_test, y_train, y_test = split_train_test(X, y) res = classifier_statistics(gnb, X_train, X_test, y_train, y_test) export_file(res, d[1], 'Naive bayes', "") def naive_bayes_balenced(): gnb = GaussianNB() for d in data: X_train, X_test, y_train, y_test, X_train_res, y_train_res = balance_dataset(d[0], 'consensus') res = classifier_statistics(gnb, X_train_res, X_test, y_train_res.ravel(), y_test.ravel()) export_file(res, d[1], 'Naive bayes balenced', "") clf_RF = RandomForestClassifier(n_estimators=800, max_depth=4) X_train, X_test, y_train, y_test, X_train_res, y_train_res = balance_dataset(data[0][0], 'consensus') res = classifier_statistics(clf_RF, X_train, X_test, y_train, y_test) pprint(res)
print_dict(clf_stats, ['predicted'])
print()
print()
diabetes = pd.read_csv('diabetes.csv')
X, y = split_dataset_transformed(diabetes, 'class')
X_train, X_test, y_train, y_test = split_train_test(X, y)
# 3.a
enc = KBinsDiscretizer(n_bins=10, encode='onehot')
X_binned = enc.fit_transform(X)
X_binned = X_binned.toarray()
X_binned_train, X_binned_test, y_binned_train, y_binned_test = split_train_test(
X_binned, y)
print('With discretization')
clf1 = RandomForestClassifier()
clf_stats = classifier_statistics(clf1, X_binned_train, X_binned_test,
y_binned_train, y_binned_test)
print_dict(clf_stats, ['predicted'])
print('Without discretization')
clf2 = RandomForestClassifier()
clf_stats = classifier_statistics(clf2, X_train, X_test, y_train, y_test)
print_dict(clf_stats, ['predicted'])
aps_train = aps_train.dropna()
aps_test = aps_test.dropna()
X_train, y_train = split_dataset(aps_train, 'class')
X_test, y_test = split_dataset(aps_train, 'class')
y_train = y_train.map({'pos': 1, 'neg': 0})
y_test = y_test.map({'pos': 1, 'neg': 0})
X_train = X_train.astype('float64')
X_test = X_test.astype('float64')
neigh = KNeighborsClassifier(n_neighbors=10)
res = classifier_statistics(neigh, X_train, X_test, y_train, y_test)
pprint(res)
"""
k=2
accuracy: 0.936
sensibility: 0.5
k=3
accuracy: 0.946
sensibility: 0.658
k=10
accuracy: 0.892
sensibility: 0.224