def k_train(clusters, dataset, labelset, verbose=False):
# Create k initial clusters and assign all points to them randomly from training data
for (label_index, feature) in enumerate(dataset):
point = Point(labelset[label_index], feature)
cluster_index = rand.randrange(len(clusters))
clusters[cluster_index].points.append(point)
# Assign cluster center randomly from training samples
for cluster in clusters:
cluster.center = cluster.points[rand.randrange(len(cluster.points))].features
# Iterate the below until convergence
mse_, mss_, ent_ = calc_kmeans_stats(clusters)
if verbose:
helper.print_stats(mse_, mss_, ent_)
counter = 0
mse_t, mss_t, ent_t = 0, 0, 0
converged = True
while update_clusters(clusters):
prev_mse, prev_mss, prev_ent = mse_t, mss_t, ent_t
mse_, mss_, ent_ = calc_kmeans_stats(clusters)
mse_t, mss_t, ent_t = mse_, mss_, ent_
if verbose:
helper.print_stats(mse_, mss_, ent_)
counter += 1
if (prev_mse - mse_t) + (prev_mss - mss_t) + (prev_ent - ent_t) < 5 and counter >= 100 and counter != 1: # Probably indicates non-convergence
converged = False
if verbose:
print("Non-convergence detected, finishing training sequence...")
break
return mse_, mss_, ent_, clusters, converged
svm) #pipeline_svm obj to be used in all svm algos
pipeline_svm_fitted = pipeline_svm.fit(docs_train.data, docs_train.target)
# svm_predict = pipeline_svm_fitted.predict(docs_test.data)
# utility.print_stats(docs_test.target, svm_predict, 'SVM Normal')
# utility.draw_roc_curve(docs_test.target, pipeline_svm_fitted.predict_proba(docs_test.data)[:, 1])
#Soft margin SVM ->
#confirm this part, not sure of any other way to implement soft margin SVM
params = {
'learning_algo__gamma': [1e-3, 1e3] #10^-3 to 10^3
}
svm_soft_margin = GridSearchCV(pipeline_svm, params, cv=5)
svm_soft_margin_fitted = svm_soft_margin.fit(docs_train.data,
docs_train.target)
svm_soft_margin_predict = svm_soft_margin_fitted.predict(docs_test.data)
utility.print_stats(docs_test.target, svm_soft_margin_predict,
'Soft Margin SVM')
utility.draw_roc_curve(
docs_test.target,
svm_soft_margin_fitted.predict_proba(docs_test.data)[:, 1])
best_params = svm_soft_margin.best_estimator_.get_params()
for param_name in sorted(params.keys()):
print("\t{}: {}".format(param_name, best_params[param_name]))
#Logistic Regression ->
# logistic_regr = LogisticRegression(penalty='l2', max_iter=5, random_state=40)
# pipeline_regr = utility.pipeline_setup(logistic_regr)
# pipeline_regr_fitted = pipeline_regr.fit(docs_train.data, docs_train.target)
# regr_predict = pipeline_regr_fitted.predict(docs_test.data)
# utility.print_stats(docs_test.target, regr_predict, 'Logistic Regression')
# utility.draw_roc_curve(docs_test.target, pipeline_regr_fitted.predict_proba(docs_test.data)[:, 1])
#SVM ->
svm = SVC(kernel='linear', probability=True, random_state=40)
pipeline_svm = utility.pipeline_setup(svm) #pipeline_svm obj to be used in all svm algos
pipeline_svm_fitted = pipeline_svm.fit(docs_train.data, docs_train.target)
# svm_predict = pipeline_svm_fitted.predict(docs_test.data)
# utility.print_stats(docs_test.target, svm_predict, 'SVM Normal')
# utility.draw_roc_curve(docs_test.target, pipeline_svm_fitted.predict_proba(docs_test.data)[:, 1])
#Soft margin SVM ->
#confirm this part, not sure of any other way to implement soft margin SVM
params = {
'learning_algo__gamma': [1e-3, 1e3] #10^-3 to 10^3
}
svm_soft_margin = GridSearchCV(pipeline_svm, params, cv=5)
svm_soft_margin_fitted = svm_soft_margin.fit(docs_train.data, docs_train.target)
svm_soft_margin_predict = svm_soft_margin_fitted.predict(docs_test.data)
utility.print_stats(docs_test.target, svm_soft_margin_predict, 'Soft Margin SVM')
utility.draw_roc_curve(docs_test.target, svm_soft_margin_fitted.predict_proba(docs_test.data)[:, 1])
best_params = svm_soft_margin.best_estimator_.get_params()
for param_name in sorted(params.keys()):
print("\t{}: {}".format(param_name, best_params[param_name]))
#Logistic Regression ->
# logistic_regr = LogisticRegression(penalty='l2', max_iter=5, random_state=40)
# pipeline_regr = utility.pipeline_setup(logistic_regr)
# pipeline_regr_fitted = pipeline_regr.fit(docs_train.data, docs_train.target)
# regr_predict = pipeline_regr_fitted.predict(docs_test.data)
# utility.print_stats(docs_test.target, regr_predict, 'Logistic Regression')
# utility.draw_roc_curve(docs_test.target, pipeline_regr_fitted.predict_proba(docs_test.data)[:, 1])
from sklearn.datasets import fetch_20newsgroups
from sklearn.naive_bayes import GaussianNB
from sklearn import metrics
import utility
categories = [
'comp.sys.ibm.pc.hardware', 'comp.sys.mac.hardware', 'misc.forsale',
'soc.religion.christian'
]
docs_train = fetch_20newsgroups(
subset='train', categories=categories, shuffle=True,
random_state=42) #, remove=('headers','footers','quotes'))
docs_test = fetch_20newsgroups(
subset='test', categories=categories, shuffle=True,
random_state=42) #, remove=('headers','footers','quotes'))
model = utility.pipeline_setup(GaussianNB())
model.fit(docs_train.data, docs_train.target)
# print(model)
# make predictions
expected = docs_test.target
predicted = model.predict(docs_test.data)
utility.print_stats(expected, predicted, 'Naive Bayes Multiclass')
from sklearn.datasets import fetch_20newsgroups import matplotlib.pyplot as pyplot from sklearn.naive_bayes import GaussianNB import utility docs_train, docs_test = utility.custom_2class_classifier() model = utility.pipeline_setup(GaussianNB()) model_fitted = model.fit(docs_train.data, docs_train.target) #print(model) # make predictions expected = docs_test.target predicted = model_fitted.predict(docs_test.data) utility.print_stats(expected, predicted, 'Naive Bayes Basic') utility.draw_roc_curve(expected, model_fitted.predict_proba(docs_test.data)[:, 1])
categories=categories,
shuffle=True,
random_state=42)
docs_test = fetch_20newsgroups(subset='test',
categories=categories,
shuffle=True,
random_state=42)
svm_basic = SVC(kernel='linear',
class_weight='balanced',
probability=True,
random_state=40)
svm_onerest = OneVsRestClassifier(svm_basic)
pipeline_svm_onerest = utility.pipeline_setup(svm_onerest)
pipeline_svm_fitted = pipeline_svm_onerest.fit(docs_train.data,
docs_train.target)
svm_predict = pipeline_svm_fitted.predict(docs_test.data)
utility.print_stats(docs_test.target, svm_predict, 'SVM OneVSOne')
svm_weighted = SVC(
kernel='linear',
class_weight='balanced',
probability=True,
random_state=40
) #balanced param to make sure both docs have same no. of samples in onevsone
svm_oneone = OneVsOneClassifier(svm_weighted)
pipeline_svm_oneone = utility.pipeline_setup(svm_oneone)
pipeline_svm_fitted = pipeline_svm_oneone.fit(docs_train.data,
docs_train.target)
svm_predict = pipeline_svm_fitted.predict(docs_test.data)
utility.print_stats(docs_test.target, svm_predict, 'SVM OneVSRest')
import utility from sklearn.datasets import fetch_20newsgroups from sklearn.svm import SVC from sklearn.multiclass import OneVsOneClassifier, OneVsRestClassifier from sklearn import metrics categories = ['comp.sys.ibm.pc.hardware', 'comp.sys.mac.hardware', 'misc.forsale', 'soc.religion.christian'] docs_train = fetch_20newsgroups(subset='train', categories=categories, shuffle=True, random_state=42) docs_test = fetch_20newsgroups(subset='test', categories=categories, shuffle=True, random_state=42) svm_basic = SVC(kernel='linear', class_weight='balanced', probability=True, random_state=40) svm_onerest = OneVsRestClassifier(svm_basic) pipeline_svm_onerest = utility.pipeline_setup(svm_onerest) pipeline_svm_fitted = pipeline_svm_onerest.fit(docs_train.data, docs_train.target) svm_predict = pipeline_svm_fitted.predict(docs_test.data) utility.print_stats(docs_test.target, svm_predict, 'SVM OneVSOne') svm_weighted = SVC(kernel='linear', class_weight='balanced', probability=True,random_state=40) #balanced param to make sure both docs have same no. of samples in onevsone svm_oneone = OneVsOneClassifier(svm_weighted) pipeline_svm_oneone = utility.pipeline_setup(svm_oneone) pipeline_svm_fitted = pipeline_svm_oneone.fit(docs_train.data, docs_train.target) svm_predict = pipeline_svm_fitted.predict(docs_test.data) utility.print_stats(docs_test.target, svm_predict, 'SVM OneVSRest')
from sklearn.datasets import fetch_20newsgroups
from sklearn.naive_bayes import GaussianNB
from sklearn import metrics
import utility
categories = ['comp.sys.ibm.pc.hardware', 'comp.sys.mac.hardware', 'misc.forsale', 'soc.religion.christian']
docs_train = fetch_20newsgroups(subset='train', categories=categories, shuffle=True, random_state=42)#, remove=('headers','footers','quotes'))
docs_test = fetch_20newsgroups(subset='test', categories=categories, shuffle=True, random_state=42)#, remove=('headers','footers','quotes'))
model = utility.pipeline_setup(GaussianNB())
model.fit(docs_train.data, docs_train.target)
# print(model)
# make predictions
expected = docs_test.target
predicted = model.predict(docs_test.data)
utility.print_stats(expected, predicted, 'Naive Bayes Multiclass')
