def loadClassifier(self, model_prefix="features"):
self.classifier = Classifier(self.data_dir,
svm_prefix=model_prefix,
has_zero_insert=self.has_zero_insert)
self.bProbability = self.classifier.probability
class CommonClassPredictor(BaseLearner):
def __init__(self, *args, **kw):
super(CommonClassPredictor, self).__init__(*args, **kw)
self.strModelPrefix = "features"
self.classifier = None
self.nan_features = []
def has_nan_features(self):
for data in self.feature_data.itervalues():
if np.any(np.isnan(data)):
return True
return False
def loadClassifier(self, model_prefix="features"):
self.classifier = Classifier(self.data_dir,
svm_prefix=model_prefix,
has_zero_insert=self.has_zero_insert)
self.bProbability = self.classifier.probability
def predict(self, aFeatureData, feature_names):
# ensurse to get the right fearues in the right order
# FIX what if NaN's are in in feature data
dctNameLookup = dict([(name, i)
for i, name in enumerate(feature_names)])
lstRequiredFeatureData = [
aFeatureData[dctNameLookup[x]] for x in self._feature_names
]
return self.classifier(lstRequiredFeatureData)
def getData(self, normalize=True):
labels = []
samples = []
for name, data in self.feature_data.iteritems():
label = self.class_labels[name]
labels += [label] * len(data)
samples += data.tolist()
labels = np.asarray(labels)
samples = np.asarray(samples)
if normalize:
lo = np.min(samples, 0)
hi = np.max(samples, 0)
# scale between -1 and +1
samples = 2.0 * (samples - lo) / (hi - lo + 0.0000001) - 1.0
# FIXME: stupid libSVM conversions
labels = map(int, labels)
samples = samples.tolist()
return labels, samples
def filter_nans(self, apply=False):
"""Find features with NA values in the data set and remove features
from the data and corresponding feature names returns the list of
removed feature names.
"""
filter_idx = np.array([], int)
feature_idx = np.arange(len(self._feature_names), dtype=int)
for data in self.feature_data.itervalues():
filter_idx = np.append(filter_idx,
feature_idx[np.any(np.isnan(data), 0)])
filter_idx = np.unique(filter_idx)
if apply:
for name in self.feature_data:
self.feature_data[name] = np.delete(self.feature_data[name],
filter_idx, 1)
if filter_idx.size > 0:
self.nan_features = self.delete_feature_names(filter_idx)
return self.nan_features
def train(self,
c,
g,
probability=True,
compensation=True,
path=None,
filename=None,
save=True):
if filename is None:
filename = splitext(self.arff_file)[0]
filename += '.model'
if path is None:
path = self.data_dir
param = svm.svm_parameter(kernel_type=svm.RBF,
C=c,
gamma=g,
probability=1 if probability else 0)
labels, samples = self.getData(normalize=True)
# because we train the SVM with dict we need to redefine the zero-insert
self.has_zero_insert = False
if not self.classifier is None:
self.classifier.setOption('hasZeroInsert', True)
if compensation:
weight, weight_label = self._calculateCompensation(labels)
param.weight = weight
param.weight_label = weight_label
param.nr_weight = len(weight)
problem = svm.svm_problem(labels, samples)
model = svm.svm_model(problem, param)
if save:
model.save(os.path.join(path, filename))
return problem, model
def exportConfusion(self, log2c, log2g, conf):
fname = join(self.data_dir,
self.arff_file.replace(".arff", ".confusion.txt"))
with open(fname, "w") as fp:
fp.write('log2(C) = %f\n' % log2c)
fp.write('log2(g) = %f\n' % log2g)
fp.write('accuracy = %f\n' % conf.ac_sample)
fp.write('\n')
fp.write('confusion matrix (absolute)\n')
fp.write('\t%s\n' %
'\t'.join([str(k) for k in self.class_names.keys()]))
for label, row in zip(self.class_names.keys(), conf.conf):
fp.write(('%d\t' % label) + '\t'.join(['%d' % i for i in row]))
fp.write('\n')
def importConfusion(self):
fname = join(self.data_dir,
self.arff_file.replace(".arff", ".confusion.txt"))
if not isfile(fname):
raise IOError("File (%s) does not exist")
with open(fname, "Ur") as fp:
log2c = float(fp.readline().split('=')[1].strip())
log2g = float(fp.readline().split('=')[1].strip())
fp.readline()
fp.readline()
fp.readline()
fp.readline()
conf_array = []
for line in fp:
line = line.strip()
if len(line) == 0:
break
items = map(int, map(float, line.split('\t')[1:]))
conf_array.append(items)
conf = ConfusionMatrix(np.asarray(conf_array))
return log2c, log2g, conf
def _calculateCompensation(self, labels):
ulabels = np.unique(labels)
count = np.bincount(labels)[ulabels]
weight = (float(len(labels)) - count) / count
weight_label = map(int, ulabels)
return weight, weight_label
def gridSearch(self,
fold=5,
c_info=None,
g_info=None,
probability=False,
compensation=True):
best_accuracy = 0
best_l2c = None
best_l2g = None
best_conf = None
n = None
for n, l2c, l2g, conf in self.iterGridSearchSVM(
c_info=c_info,
g_info=g_info,
fold=fold,
probability=probability,
compensation=compensation):
accuracy = conf.ac_sample
if accuracy > best_accuracy:
best_accuracy = accuracy
best_l2c = l2c
best_l2g = l2g
best_conf = conf
return n, best_l2c, best_l2g, best_conf
def iterGridSearchSVM(self,
c_info=None,
g_info=None,
fold=5,
probability=False,
compensation=True):
swap = lambda a, b: (b, a)
if not c_info is None and len(c_info) >= 3:
c_begin, c_end, c_step = c_info[:3]
else:
c_begin, c_end, c_step = -5, 15, 2
if c_end < c_begin:
c_begin, c_end = swap(c_begin, c_end)
c_step = abs(c_step)
if not g_info is None and len(g_info) >= 3:
g_begin, g_end, g_step = g_info[:3]
else:
g_begin, g_end, g_step = -15, 3, 2
if g_end < g_begin:
g_begin, g_end = swap(g_begin, g_end)
g_step = abs(g_step)
labels, samples = self.getData(normalize=True)
problem = svm.svm_problem(labels, samples)
if compensation:
weight, weight_label = self._calculateCompensation(labels)
n = (c_end - c_begin) / c_step + 1
n *= (g_end - g_begin) / g_step + 1
l2c = c_begin
while l2c <= c_end:
l2g = g_begin
while l2g <= g_end:
param = svm.svm_parameter(kernel_type=svm.RBF,
C=2.**l2c,
gamma=2.**l2g,
probability=1 if probability else 0)
if compensation:
param.weight = weight
param.weight_label = weight_label
param.nr_weight = len(weight)
predictions = svm.cross_validation(problem, param, fold)
predictions = map(int, predictions)
conf = ConfusionMatrix.from_lists(labels, predictions,
self.class_names.keys())
yield n, l2c, l2g, conf
l2g += g_step
l2c += c_step
class CommonClassPredictor(BaseLearner):
def __init__(self, *args, **kw):
super(CommonClassPredictor, self).__init__(*args, **kw)
self.strModelPrefix = "features"
self.classifier = None
self.nan_features = []
def has_nan_features(self):
for data in self.feature_data.itervalues():
if np.any(np.isnan(data)):
return True
return False
def loadClassifier(self, model_prefix="features"):
self.classifier = Classifier(self.data_dir, svm_prefix=model_prefix, has_zero_insert=self.has_zero_insert)
self.bProbability = self.classifier.probability
def predict(self, aFeatureData, feature_names):
# ensurse to get the right fearues in the right order
# FIX what if NaN's are in in feature data
dctNameLookup = dict([(name, i) for i, name in enumerate(feature_names)])
lstRequiredFeatureData = [aFeatureData[dctNameLookup[x]] for x in self._feature_names]
return self.classifier(lstRequiredFeatureData)
def getData(self, normalize=True):
labels = []
samples = []
for name, data in self.feature_data.iteritems():
label = self.class_labels[name]
labels += [label] * len(data)
samples += data.tolist()
labels = np.asarray(labels)
samples = np.asarray(samples)
if normalize:
lo = np.min(samples, 0)
hi = np.max(samples, 0)
# scale between -1 and +1
samples = 2.0 * (samples - lo) / (hi - lo + 0.0000001) - 1.0
# FIXME: stupid libSVM conversions
labels = map(int, labels)
samples = samples.tolist()
return labels, samples
def filter_nans(self, apply=False):
"""Find features with NA values in the data set and remove features
from the data and corresponding feature names returns the list of
removed feature names.
"""
filter_idx = np.array([], int)
feature_idx = np.arange(len(self._feature_names), dtype=int)
for data in self.feature_data.itervalues():
filter_idx = np.append(filter_idx, feature_idx[np.any(np.isnan(data), 0)])
filter_idx = np.unique(filter_idx)
if apply:
for name in self.feature_data:
self.feature_data[name] = np.delete(self.feature_data[name], filter_idx, 1)
if filter_idx.size > 0:
self.nan_features = self.delete_feature_names(filter_idx)
return self.nan_features
def train(self, c, g, probability=True, compensation=True, path=None, filename=None, save=True):
if filename is None:
filename = splitext(self.arff_file)[0]
filename += ".model"
if path is None:
path = self.data_dir
param = svm.svm_parameter(kernel_type=svm.RBF, C=c, gamma=g, probability=1 if probability else 0)
labels, samples = self.getData(normalize=True)
# because we train the SVM with dict we need to redefine the zero-insert
self.has_zero_insert = False
if not self.classifier is None:
self.classifier.setOption("hasZeroInsert", True)
if compensation:
weight, weight_label = self._calculateCompensation(labels)
param.weight = weight
param.weight_label = weight_label
param.nr_weight = len(weight)
problem = svm.svm_problem(labels, samples)
model = svm.svm_model(problem, param)
if save:
model.save(os.path.join(path, filename))
return problem, model
def exportConfusion(self, log2c, log2g, conf):
fname = join(self.data_dir, self.arff_file.replace(".arff", ".confusion.txt"))
with open(fname, "w") as fp:
fp.write("log2(C) = %f\n" % log2c)
fp.write("log2(g) = %f\n" % log2g)
fp.write("accuracy = %f\n" % conf.ac_sample)
fp.write("\n")
fp.write("confusion matrix (absolute)\n")
fp.write("\t%s\n" % "\t".join([str(k) for k in self.class_names.keys()]))
for label, row in zip(self.class_names.keys(), conf.conf):
fp.write(("%d\t" % label) + "\t".join(["%d" % i for i in row]))
fp.write("\n")
def importConfusion(self):
fname = join(self.data_dir, self.arff_file.replace(".arff", ".confusion.txt"))
if not isfile(fname):
raise IOError("File (%s) does not exist")
with open(fname, "Ur") as fp:
log2c = float(fp.readline().split("=")[1].strip())
log2g = float(fp.readline().split("=")[1].strip())
fp.readline()
fp.readline()
fp.readline()
fp.readline()
conf_array = []
for line in fp:
line = line.strip()
if len(line) == 0:
break
items = map(int, map(float, line.split("\t")[1:]))
conf_array.append(items)
conf = ConfusionMatrix(np.asarray(conf_array))
return log2c, log2g, conf
def _calculateCompensation(self, labels):
ulabels = np.unique(labels)
count = np.bincount(labels)[ulabels]
weight = (float(len(labels)) - count) / count
weight_label = map(int, ulabels)
return weight, weight_label
def gridSearch(self, fold=5, c_info=None, g_info=None, probability=False, compensation=True):
best_accuracy = 0
best_l2c = None
best_l2g = None
best_conf = None
n = None
for n, l2c, l2g, conf in self.iterGridSearchSVM(
c_info=c_info, g_info=g_info, fold=fold, probability=probability, compensation=compensation
):
accuracy = conf.ac_sample
if accuracy > best_accuracy:
best_accuracy = accuracy
best_l2c = l2c
best_l2g = l2g
best_conf = conf
return n, best_l2c, best_l2g, best_conf
def iterGridSearchSVM(self, c_info=None, g_info=None, fold=5, probability=False, compensation=True):
swap = lambda a, b: (b, a)
if not c_info is None and len(c_info) >= 3:
c_begin, c_end, c_step = c_info[:3]
else:
c_begin, c_end, c_step = -5, 15, 2
if c_end < c_begin:
c_begin, c_end = swap(c_begin, c_end)
c_step = abs(c_step)
if not g_info is None and len(g_info) >= 3:
g_begin, g_end, g_step = g_info[:3]
else:
g_begin, g_end, g_step = -15, 3, 2
if g_end < g_begin:
g_begin, g_end = swap(g_begin, g_end)
g_step = abs(g_step)
labels, samples = self.getData(normalize=True)
problem = svm.svm_problem(labels, samples)
if compensation:
weight, weight_label = self._calculateCompensation(labels)
n = (c_end - c_begin) / c_step + 1
n *= (g_end - g_begin) / g_step + 1
l2c = c_begin
while l2c <= c_end:
l2g = g_begin
while l2g <= g_end:
param = svm.svm_parameter(
kernel_type=svm.RBF, C=2.0 ** l2c, gamma=2.0 ** l2g, probability=1 if probability else 0
)
if compensation:
param.weight = weight
param.weight_label = weight_label
param.nr_weight = len(weight)
predictions = svm.cross_validation(problem, param, fold)
predictions = map(int, predictions)
conf = ConfusionMatrix.from_lists(labels, predictions, self.class_names.keys())
yield n, l2c, l2g, conf
l2g += g_step
l2c += c_step
def loadClassifier(self, model_prefix="features"):
self.classifier = Classifier(self.data_dir, svm_prefix=model_prefix, has_zero_insert=self.has_zero_insert)
self.bProbability = self.classifier.probability
def loadClassifier(self, model_prefix="features"):
self.classifier = Classifier(self.data_dir, self.logger,
strSvmPrefix=model_prefix,
hasZeroInsert=self.has_zero_insert)
self.bProbability = self.classifier.bProbability