def bench(train_data, test_data, n_trees):
"""
Benchmark Extratrees
Trains a ExtraForest with `n_trees` trees and returns training time,
evaluation time and testing accuracy.
Args:
train_data (extratrees.Datset): Train set
test_data (extratrees.Datset): Test set
n_trees (int): Number of trees
"""
print('-' * 20, "This implementation")
print("Going to train %d trees" % n_trees)
forest = ExtraForest(n_trees=n_trees, n_min=10, criterion="entropy")
print("Training...")
t_start = time.time()
forest.fit(train_data)
t_train = time.time() - t_start
print("Trained in %.3fs" % t_train)
print("Testing")
t_start = time.time()
test_score = score(forest, test_data)
t_test = time.time() - t_start
print("Evaluated in %.3fs" % t_test)
print("Score: %.3f" % test_score)
def test_fit(self):
forest = ExtraForest(n_trees=3)
forest.fit(self.fourths)
self.assertTrue(forest._is_classifier)
self.assertTrue(len(forest.trees), 3)
for tree in forest.trees:
self.assertIsInstance(tree, ExtraTree)
self.assertTrue(tree._is_classifier)
def test_predict_forest_classify(self):
random.seed(0)
forest = ExtraForest(n_trees=3, n_min=1)
forest.fit(self.fourths)
pred = forest.predict(self.fourths[0])
for tree in forest.trees:
self.assertEqual(tree.root_node.left, [0.75, 0.25])
self.assertEqual(tree.root_node.right, [0.25, 0.75])
self.assertEqual(pred, [1, 0, 0, 0, 0, 1, 1, 1])
def test_predict_forest_regression(self):
xdata = ([0], [0], [0], [0], [1], [1], [1], [1])
ydata = (2.0, 2.1, 2.2, 2.3, 3.0, 3.1, 3.2, 3.3)
forest = ExtraForest()
forest.fit((xdata, ydata))
pred = forest.predict(xdata)
# These are the means of the xdata classes, since after the first split
# the attributes are uniform
for zval in pred[:4]:
self.assertAlmostEqual(zval, 2.15)
for zval in pred[4:]:
self.assertAlmostEqual(zval, 3.15)
def test_predict(self):
random.seed(0)
forest = ExtraForest(n_trees=3, n_min=1)
forest.fit(self.fourths)
pred = forest.predict(self.fourths[0])
# For all trees, the first split will be somewhere inbetween the single
# attributes of `fourths[0`. There will be no second split as
# the attributes are uniform after the first split.
for tree in forest.trees:
self.assertEqual(tree.root_node.left, [0.75, 0.25])
self.assertEqual(tree.root_node.right, [0.25, 0.75])
self.assertEqual(pred, [1, 0, 0, 0, 0, 1, 1, 1])
def test_init_forest(self):
forest = ExtraForest(n_trees=5, k_value=3, n_min=1)
self.assertEqual(forest.n_trees, 5)
self.assertEqual(forest.k_value, 3)
self.assertEqual(forest.n_min, 1)
self.assertFalse(forest._is_classifier)
self.assertEqual(forest.trees, [])
def get_train_time(train_data):
"""
Benchmark Extratrees
Trains a ExtraTree trees and returns training time.
Args:
train_data (extratrees.Datset): Train set
"""
forest = ExtraForest(n_trees=10, n_min=10, criterion="entropy")
t_start = time.time()
forest.fit(train_data)
t_train = (time.time() - t_start) / 10
return t_train
def test_repr_no_fit(self):
forest = ExtraForest()
self.assertEqual(str(forest), "<Forest (10 trees), not fitted>")
""" compare.py - Compare training time of extratrees and sklearn """
import time
import matplotlib.pyplot as plt
from sklearn.ensemble import ExtraTreesClassifier
from src.extratrees import ExtraForest
from docs.helpers import load_data, TRAIN_FILE, TEST_FILE
if __name__ == '__main__':
TRAIN_SET = load_data(TRAIN_FILE)
TEST_SET = load_data(TEST_FILE)
this_clf = ExtraForest(n_trees=1, n_min=10, criterion="gini")
skl_clf = ExtraTreesClassifier(criterion="gini", min_samples_split=10)
sizes = (100, 1000, 2000, 5000, 10000, 20000, 50000)
times_this = [0] * len(sizes)
times_skl = [0] * len(sizes)
for idx, size in enumerate(sizes):
# This clf
t0 = time.time()
this_clf.fit((TRAIN_SET[0][:size], TRAIN_SET[1][:size]))
times_this[idx] = time.time() - t0
# Sklearn
t0 = time.time()
skl_clf.fit(TRAIN_SET[0][:size], TRAIN_SET[1][:size])
grid_pred = classifier.predict([[x, y]])
col = (grid_pred[0]), 0, (1 - grid_pred[0])
ax.add_patch(
patches.Rectangle((x, y), (2 / gridsize), (2 / gridsize),
facecolor=col))
# Create data
data = create_dataset()
# Train tree
tree = ExtraTree(n_min=5)
tree.fit(data)
# Train forest
forest = ExtraForest(n_trees=10, n_min=5)
forest.fit(data)
# Predict
pred_tree = tree.predict(data[0])
pred_forest = forest.predict(data[0])
fig, axes = plt.subplots(2, 3)
# Plot data
axes[0][0].set_title('True')
for coords, cls in zip(*data):
axes[0][0].plot(*coords, marker='o', color=colorize_class(cls))
# Plot tree result
axes[0][1].set_title("Single tree\n (min_samples=5, k=*)")