class Koala(object):
data = None
classifier = None
train_kwargs = None
_mc = None
def __init__(self,data=None, target=None, train=False):
if data is not None:
self.set_data(data)
if target is not None:
self.set_target(target)
if train:
self.train(test_size=0.1)
return None
def save(self, path):
odata = {'data': self.data, 'mc': self._mc, 'classifier': self.classifier}
with open(path, 'wb') as f:
pickle_save(odata, f)
def load(self, path):
with open(path, 'rb') as f:
try:
odata = pickle_load(f)
self.data = odata['data']
self._mc = odata['mc']
self.classifier = odata['classifier']
except Exception as e:
print("Invalid data: %s" % (str(e)))
def set_data(self, df):
self.data = Dataset(df)
return None
def set_target(self, column):
self.data.role[column] = self.data.TARGET
return None
def train(self, **kwargs):
self.train_kwargs = kwargs
test_size = 0.1
if kwargs.get('test_size') is not None:
test_size = kwargs.get('test_size')
del kwargs['test_size']
self._mc = ModelComparison()
self._mc.train_test_split(self.data, test_size=test_size, random_state=randint(2**16))
self._mc['RFC'] = RandomForestClassifier(**kwargs)
self.classifier = self._mc['RFC']
self._mc.fit()
return None
def predict(self, X):
try:
predictions = self._mc.le.inverse_transform(self._mc['RFC'].predict(X))
except AttributeError:
predictions = self._mc['RFC'].predict(X)
return predictions
def accuracy_score(self):
return self._mc.accuracy_score()[0]
def precision_score(self, **kwargs):
return self._mc.precision_score(**kwargs)[0]
def recall_score(self, **kwargs):
return self._mc.recall_score(**kwargs)[0]
def f1_score(self, **kwargs):
return self._mc.f1_score(**kwargs)[0]
def metric_score(self, metric, **kwargs):
return self._mc.metric(metric, **kwargs)[0]
def confusion_matrix(self):
return self._mc.cm('RFC')
def feature_importance(self):
rf_weights = list(self._mc['RFC'].feature_importances_)
rf_inputs = list(self.data.filter_cols(role=self.data.INPUT))
return DataFrame(data=rf_weights, index=rf_inputs,columns=['weight']).sort('weight', ascending=False)
def feature_reduction_scores(self):
K = 1 # to be developped more later on
ordered_features = list(self.feature_importance().index)[::-1]
ordered_features.append(self.data.filter_cols(role=self.data.TARGET)[0])
idx = []
accuracy_score = []
for i in range(len(ordered_features[:-1])):
accuracy_score_i = []
idx.append(len(ordered_features[i:])-1)
for _ in range(K):
kl = Koala(data=self.data.frame[ordered_features[i:]], target=self.data.filter_cols(role=self.data.TARGET)[0])
kl.train(**self.train_kwargs)
accuracy_score_i.append(kl.accuracy_score())
accuracy_score.append(sum(accuracy_score_i) / len(accuracy_score_i))
scores = DataFrame(data={'accuracy': accuracy_score}, index=idx)
return scores
class Koala(object):
data = None
classifier = None
train_kwargs = None
_mc = None
def __init__(self, data=None, target=None, train=False):
if data is not None:
self.set_data(data)
if target is not None:
self.set_target(target)
if train:
self.train(test_size=0.1)
return None
def save(self, path):
odata = {
'data': self.data,
'mc': self._mc,
'classifier': self.classifier
}
with open(path, 'wb') as f:
pickle_save(odata, f)
def load(self, path):
with open(path, 'rb') as f:
try:
odata = pickle_load(f)
self.data = odata['data']
self._mc = odata['mc']
self.classifier = odata['classifier']
except Exception as e:
print("Invalid data: %s" % (str(e)))
def set_data(self, df):
self.data = Dataset(df)
return None
def set_target(self, column):
self.data.role[column] = self.data.TARGET
return None
def train(self, **kwargs):
self.train_kwargs = kwargs
test_size = 0.1
if kwargs.get('test_size') is not None:
test_size = kwargs.get('test_size')
del kwargs['test_size']
self._mc = ModelComparison()
self._mc.train_test_split(self.data,
test_size=test_size,
random_state=randint(2**16))
self._mc['RFC'] = RandomForestClassifier(**kwargs)
self.classifier = self._mc['RFC']
self._mc.fit()
return None
def predict(self, X):
try:
predictions = self._mc.le.inverse_transform(
self._mc['RFC'].predict(X))
except AttributeError:
predictions = self._mc['RFC'].predict(X)
return predictions
def accuracy_score(self):
return self._mc.accuracy_score()[0]
def precision_score(self, **kwargs):
return self._mc.precision_score(**kwargs)[0]
def recall_score(self, **kwargs):
return self._mc.recall_score(**kwargs)[0]
def f1_score(self, **kwargs):
return self._mc.f1_score(**kwargs)[0]
def metric_score(self, metric, **kwargs):
return self._mc.metric(metric, **kwargs)[0]
def confusion_matrix(self):
return self._mc.cm('RFC')
def feature_importance(self):
rf_weights = list(self._mc['RFC'].feature_importances_)
rf_inputs = list(self.data.filter_cols(role=self.data.INPUT))
return DataFrame(data=rf_weights, index=rf_inputs,
columns=['weight']).sort('weight', ascending=False)
def feature_reduction_scores(self):
K = 1 # to be developped more later on
ordered_features = list(self.feature_importance().index)[::-1]
ordered_features.append(
self.data.filter_cols(role=self.data.TARGET)[0])
idx = []
accuracy_score = []
for i in range(len(ordered_features[:-1])):
accuracy_score_i = []
idx.append(len(ordered_features[i:]) - 1)
for _ in range(K):
kl = Koala(
data=self.data.frame[ordered_features[i:]],
target=self.data.filter_cols(role=self.data.TARGET)[0])
kl.train(**self.train_kwargs)
accuracy_score_i.append(kl.accuracy_score())
accuracy_score.append(
sum(accuracy_score_i) / len(accuracy_score_i))
scores = DataFrame(data={'accuracy': accuracy_score}, index=idx)
return scores
def set_data(self, df):
self.data = Dataset(df)
return None
def set_data(self, df):
self.data = Dataset(df)
return None