def model_mil(train_bags, train_labels, test_bags, test_labels):
start_time = datetime.now()
classifiers = {}
classifiers['MissSVM'] = misvm.MissSVM(kernel='linear',
C=1.0,
max_iters=20)
classifiers['sbMIL'] = misvm.sbMIL(kernel='linear', eta=0.1, C=1e2)
classifiers['SIL'] = misvm.SIL(kernel='linear', C=1.0)
# Train/Evaluate classifiers
perf = {}
for algorithm, classifier in classifiers.items():
classifier.fit(train_bags, train_labels)
predictions = classifier.predict(test_bags)
if algorithm == 'sbMIL' or algorithm == 'MissSVM':
predictions = np.sign(predictions) + 1
[[TN, FP], [FN, TP]] = confusion_matrix(test_labels, predictions)
FDR = TP / (TP + FN)
FAR = FP / (TN + FP)
perf[algorithm] = {'FDR': FDR, 'FAR': FAR}
print '[%s]algorithm %s done...' % (time.asctime(
time.localtime(time.time())), algorithm)
print '[%s]%s done for %d seconds...' % (time.asctime(
time.localtime(time.time())), sys._getframe().f_code.co_name,
(datetime.now() -
start_time).seconds)
return perf
def main():
# Load list of C4.5 Examples
example_set = parse_c45('musk1')
table = re.findall(r"<(.*)>", str(example_set))
output.write(str(example_set))
print(len(table))
print(table)
# Get stats to normalize data
raw_data = np.array(example_set.to_float())
data_mean = np.average(raw_data, axis=0)
data_std = np.std(raw_data, axis=0)
data_std[np.nonzero(data_std == 0.0)] = 1.0
def normalizer(ex):
ex = np.array(ex)
normed = ((ex - data_mean) / data_std)
# The ...[:, 2:-1] removes first two columns and last column,
# which are the bag/instance ids and class label, as part of the
# normalization process
return normed[2:-1]
# Group examples into bags
bagset = bag_set(example_set)
# Convert bags to NumPy arrays
bags = [np.array(b.to_float(normalizer)) for b in bagset]
labels = np.array([b.label for b in bagset], dtype=float)
# Convert 0/1 labels to -1/1 labels
labels = 2 * labels - 1
# Spilt dataset arbitrarily to train/test sets
train_bags = bags[10:]
train_labels = labels[10:]
test_bags = bags[:10]
test_labels = labels[:10]
# Construct classifiers
classifiers = {}
classifiers['MissSVM'] = misvm.MissSVM(kernel='linear',
C=1.0,
max_iters=20)
classifiers['sbMIL'] = misvm.sbMIL(kernel='linear', eta=0.1, C=1e2)
classifiers['SIL'] = misvm.SIL(kernel='linear', C=1.0)
# Train/Evaluate classifiers
accuracies = {}
for algorithm, classifier in classifiers.items():
classifier.fit(train_bags, train_labels)
predictions = classifier.predict(test_bags)
accuracies[algorithm] = np.average(test_labels == np.sign(predictions))
for algorithm, accuracy in accuracies.items():
print('\n%s Accuracy: %.1f%%' % (algorithm, 100 * accuracy))
output.write('\n%s Accuracy: %.1f%%' % (algorithm, 100 * accuracy))
def train(self, dataset, labels):
self.SVMs = []
index = 0
glob_start = time.time()
for label in np.transpose(labels):
start = time.time()
classifier = misvm.SIL(kernel='linear', C=1.0)
# classifier = misvm.MISVM(kernel='linear', C=1.0, max_iters=20)
# classifier = misvm.miSVM(kernel='linear', C=1.0, max_iters=20)
classifier.fit(dataset, np.array(label))
self.SVMs.append(classifier)
index += 1
print "Trained " + str(index) + " of " + str(
len(np.transpose(labels))) + " in " + str(
int(time.time()) - int(start)) + " sec"
print "It took " + str(
(int(time.time()) - int(glob_start)) / 60) + " minutes"
def misvm_classify(verbose, output, vectors, labels):
# perform the actual misvm classification
if verbose:
print "Creating train and test bags and labels..."
bags = [numpy.array(vectors[v], dtype=float)
for v in vectors] # numpy-format matrix for use in misvm
labels = numpy.array([labels[l] for l in labels],
dtype=float) # numpy-format labels for use in misvm
# Spilt dataset into train and test sets
train_bags = []
train_labels = []
test_bags = []
test_labels = []
for i in range(len(labels)):
if i % 2 == 0:
train_bags.append(bags[i])
train_labels.append(labels[i])
else:
test_bags.append(bags[i])
test_labels.append(labels[i])
if verbose:
print "MISVM Classifying..."
if output != 'NONE':
sys.stdout = open(output, 'w')
# establish classifiers
classifiers = {
'sbMIL': misvm.sbMIL(kernel='rbf', eta=0.1, C=1.0),
'SIL': misvm.SIL(kernel='rbf', C=1.0),
'MISVM': misvm.MISVM(kernel='rbf', C=1.0, max_iters=100),
}
# Train/Evaluate classifiers
accuracies = {}
for algorithm, classifier in classifiers.items():
classifier.fit(train_bags, train_labels)
predictions = classifier.predict(test_bags)
accuracies[algorithm] = numpy.average(
test_labels == numpy.sign(predictions))
for algorithm, accuracy in accuracies.items():
print '\n%s Accuracy: %.1f%%' % (algorithm, 100 * accuracy)
if output != 'NONE':
sys.stdout = sys.__stdout__ # reset stdout to normal
def main():
# Load list of C4.5 Examples
example_set = parse_c45('musk1')
# Group examples into bags
bagset = bag_set(example_set)
# Convert bags to NumPy arrays
# (The ...[:, 2:-1] removes first two columns and last column,
# which are the bag/instance ids and class label)
bags = [np.array(b.to_float())[:, 2:-1] for b in bagset]
labels = np.array([b.label for b in bagset], dtype=float)
# Convert 0/1 labels to -1/1 labels
labels = 2 * labels - 1
# Spilt dataset arbitrarily to train/test sets
train_bags = bags[10:]
train_labels = labels[10:]
test_bags = bags[:10]
test_labels = labels[:10]
# Construct classifiers
classifiers = {}
classifiers['MissSVM'] = misvm.MissSVM(kernel='linear',
C=1.0,
max_iters=10)
classifiers['sbMIL'] = misvm.sbMIL(kernel='linear', eta=0.1, C=1.0)
classifiers['SIL'] = misvm.SIL(kernel='linear', C=1.0)
# Train/Evaluate classifiers
accuracies = {}
for algorithm, classifier in classifiers.items():
classifier.fit(train_bags, train_labels)
predictions = classifier.predict(test_bags)
accuracies[algorithm] = np.average(test_labels == np.sign(predictions))
for algorithm, accuracy in accuracies.items():
print('\n%s Accuracy: %.1f%%' % (algorithm, 100 * accuracy))