def test_split(self):
entire_range = range(0, 561)
test_beginning = split(entire_range, 0, 5)
self.assertEquals(test_beginning[0], range(112, 561))
self.assertEquals(test_beginning[1], range(0, 112))
test_middle = split(entire_range, 2, 5)
self.assertEquals(test_middle[0], range(0, 224) + range(336, 561))
self.assertEquals(test_middle[1], range(224, 336))
test_end = split(entire_range, 4, 5)
self.assertEquals(test_end[0], range(0,448))
self.assertEquals(test_end[1], range(448,561))
def main():
train_data = diversity_experiment.get_nursery_data()
shuffle(train_data.examples)
outcomes = {'unpruned': [],
'conservative': [],
'liberal': []}
for i in range(0, K):
# Split into training, test, and validation sets
training_set, the_rest = diversity_experiment.split(train_data.examples, i, K)
test_set, validation_set = validation_split(the_rest)
# Learn the original tree
unpruned = decision_tree_builder.learn_tree(train_data.features,
train_data.class_vals,
DIV_FUNCT,
training_set)
# Make pruned versions too
print "Pruning conservatively"
conservative_tree = TreePruner(copy.deepcopy(unpruned), 'conservative', validation_set).root
print "Pruning liberally."
liberal_tree = TreePruner(copy.deepcopy(unpruned), 'liberal', validation_set).root
trees = {
'unpruned': unpruned,
'liberal': liberal_tree,
'conservative': conservative_tree
}
# Test ALL the trees!
for tree in trees:
outcomes[tree].append(evaluate(trees[tree], test_set))
pairwise = [
('unpruned', 'liberal'),
('unpruned', 'conservative'),
('liberal', 'conservative')
]
# Get t-statistics
diversity_experiment.analyze(outcomes, pairwise, [], 'prune.csv', K)
def main():
train_data = diversity_experiment.get_nursery_data()
shuffle(train_data.examples)
outcomes = {'unpruned': [], 'conservative': [], 'liberal': []}
for i in range(0, K):
# Split into training, test, and validation sets
training_set, the_rest = diversity_experiment.split(
train_data.examples, i, K)
test_set, validation_set = validation_split(the_rest)
# Learn the original tree
unpruned = decision_tree_builder.learn_tree(train_data.features,
train_data.class_vals,
DIV_FUNCT, training_set)
# Make pruned versions too
print "Pruning conservatively"
conservative_tree = TreePruner(copy.deepcopy(unpruned), 'conservative',
validation_set).root
print "Pruning liberally."
liberal_tree = TreePruner(copy.deepcopy(unpruned), 'liberal',
validation_set).root
trees = {
'unpruned': unpruned,
'liberal': liberal_tree,
'conservative': conservative_tree
}
# Test ALL the trees!
for tree in trees:
outcomes[tree].append(evaluate(trees[tree], test_set))
pairwise = [('unpruned', 'liberal'), ('unpruned', 'conservative'),
('liberal', 'conservative')]
# Get t-statistics
diversity_experiment.analyze(outcomes, pairwise, [], 'prune.csv', K)