def train(self, trainset):
"""
Trains a random forest using RT-Rank.
"""
self.n_classes = len(trainset.metadata['targets'])
self.input_size = trainset.metadata['input_size']
if self.max_height is None or self.max_height < 1:
maxdepth = -1
else:
maxdepth = self.max_height
if self.n_features_per_node is None:
k_random_features = int(np.round(np.sqrt(self.input_size)))
else:
k_random_features = self.n_features_per_node
self.trees = pyrtrank.pyrtrank_interface(self.input_size, maxdepth,
k_random_features, self.seed)
# Add examples to RT-Rank training set
import sys
print 'Passing training data to RT-Rank library'
sys.stdout.flush()
for x, y in trainset:
self.trees.add_example(float(y), 1, [xi for xi in x])
print 'Training decision trees'
sys.stdout.flush()
for t in range(self.n_trees):
# Add trees one at a time
self.trees.train(True, True, True, self.zeros_are_missing, 1)
def train(self,trainset):
"""
Trains an ensemble of trees using gradient boosting and RT-Rank.
"""
self.input_size = trainset.metadata['input_size']
if self.max_height is None or self.max_height < 1:
maxdepth = -1
else:
maxdepth = self.max_height
if self.n_features_per_node is None:
k_random_features = int(np.round(np.sqrt(self.input_size)))
else:
k_random_features = self.n_features_per_node
self.trees = pyrtrank.pyrtrank_interface(self.input_size,
maxdepth,
k_random_features,
self.seed)
# Add examples to RT-Rank training set
import sys
print 'Passing training data to RT-Rank library'
sys.stdout.flush()
for x,y in trainset:
self.trees.add_example(float(y),1,[xi for xi in x])
print 'Training decision trees'
sys.stdout.flush()
for tree in range(self.n_trees):
# Add trees one at a time
self.trees.train(False,False,False,self.zeros_are_missing,self.n_processors)
self.trees.subtract_predictions_from_targets(tree,self.learning_rate)
def train(self,trainset):
"""
Trains a random forest using RT-Rank.
"""
self.n_classes = len(trainset.metadata['targets'])
self.input_size = trainset.metadata['input_size']
if self.max_height is None or self.max_height < 1:
maxdepth = -1
else:
maxdepth = self.max_height
if self.n_features_per_node is None:
k_random_features = int(np.round(np.sqrt(self.input_size)))
else:
k_random_features = self.n_features_per_node
self.trees = pyrtrank.pyrtrank_interface(self.input_size,
maxdepth,
k_random_features,
self.seed)
# Add examples to RT-Rank training set
import sys
print 'Passing training data to RT-Rank library'
sys.stdout.flush()
for x,y in trainset:
self.trees.add_example(float(y),1,[xi for xi in x])
print 'Training decision trees'
sys.stdout.flush()
for t in range(self.n_trees):
# Add trees one at a time
self.trees.train(True,True,True,self.zeros_are_missing,1)
def train(self, trainset):
"""
Trains an ensemble of trees using gradient boosting and RT-Rank.
"""
self.input_size = trainset.metadata['input_size']
if self.max_height is None or self.max_height < 1:
maxdepth = -1
else:
maxdepth = self.max_height
if self.n_features_per_node is None:
k_random_features = int(np.round(np.sqrt(self.input_size)))
else:
k_random_features = self.n_features_per_node
self.trees = pyrtrank.pyrtrank_interface(self.input_size, maxdepth,
k_random_features, self.seed)
# Add examples to RT-Rank training set
import sys
print 'Passing training data to RT-Rank library'
sys.stdout.flush()
for x, y in trainset:
self.trees.add_example(float(y), 1, [xi for xi in x])
print 'Training decision trees'
sys.stdout.flush()
for tree in range(self.n_trees):
# Add trees one at a time
self.trees.train(False, False, False, self.zeros_are_missing,
self.n_processors)
self.trees.subtract_predictions_from_targets(
tree, self.learning_rate)