def test_octavo(
num_points,
classes,
xbound,
ybound,
zbound,
max_depth,
min_node_size,
min_loss,
expected,
):
xy_parent = data_for_tests.make_octavo(
num_points, classes, xbound, ybound, zbound
).values
X = xy_parent[:, :-1]
y = xy_parent[:, -1]
forest = random_forest.grow_random_forest(
X, y, num_trees=20, max_features=2, min_node_size=1
)
predictions = random_forest.forest_predict(forest, X)
targets = y
tfpns = evaluation.tfpn(predictions, targets)
cm = evaluation.make_confusion_matrix(*tfpns, percentage=True)
result = np.array(
[evaluation.precision(cm), evaluation.sensitivity(cm), evaluation.fpr(cm)]
)
expected = np.array(expected)
assert np.any(np.abs(expected - result) < 0.01)
def test_random_forest_accuracy():
X, t = load_wine(return_X_y=True)
X_train, X_test, t_train, t_test = util.split_data(X, t, seed=0)
forest = rf.random_forest(X_train, t_train)
predictions = rf.forest_predict(X_test, forest)
count = 0
for i in range(len(t_test)):
if (predictions[i] == t_test[i]):
count += 1
rf_score = count / len(t_test)
assert (rf_score >
0.8), ('Prediction of bagged trees not good enough with ',
rf_score, ' accuracy.')
def test_diagonal_ndim(num_points, dim, max_features, expected, precision_bound):
xy_parent = data_for_tests.make_diagonal_ndim(num_points, dim).values
X = xy_parent[:, :-1]
y = xy_parent[:, -1]
forest = random_forest.grow_random_forest(
X, y, num_trees=30, max_depth=20, max_features=max_features, min_node_size=1
)
predictions = random_forest.forest_predict(forest, X)
targets = y
tfpns = evaluation.tfpn(predictions, targets)
cm = evaluation.make_confusion_matrix(*tfpns, percentage=True)
result = np.array(
[evaluation.precision(cm), evaluation.sensitivity(cm), evaluation.fpr(cm)]
)
expected = np.array(expected)
print(precision_bound)
assert np.any(np.abs(expected - result) < precision_bound)
def test_random_forest_cv_accuracy():
X, t = load_wine(return_X_y=True)
cv_data = util.crossvalidate_data(X, t, seed=0)
K = len(cv_data)
rf_score = 0
for i in range(K):
X_train = cv_data[i][0]
t_train = cv_data[i][1]
X_test = cv_data[i][2]
t_test = cv_data[i][3]
forest = rf.random_forest(X_train, t_train)
predictions = rf.forest_predict(X_test, forest)
count = 0
for i in range(len(t_test)):
if (predictions[i] == t_test[i]):
count += 1
rf_score += count / len(t_test)
rf_score = rf_score / 5
assert (rf_score >
0.8), ('Prediction of bagged trees not good enough with ',
rf_score, ' accuracy.')
import data_for_tests
num_points, dim, max_features = 1000, 5, 2
# generate data
xy_parent = data_for_tests.make_diagonal_ndim(num_points, dim).values
X = xy_parent[:, :-1]
y = xy_parent[:, -1]
# train the model -- grow the forest
forest = random_forest.grow_random_forest(
X, y, num_trees=30, max_depth=20, max_features=max_features, min_node_size=1
)
# make predictions
predictions = random_forest.forest_predict(forest, X)
# calculate the numbers of true positives, false positives, true negatives, false negatives
tfpns = evaluation.tfpn(predictions, y)
# calculate the confusion matrix
cm = evaluation.make_confusion_matrix(*tfpns, percentage=True)
# calculate metrics: precision, sensitivity, false-positive-rate
metrics = np.array(
[evaluation.precision(cm), evaluation.sensitivity(cm), evaluation.fpr(cm)]
)
print(
f"{num_points} points are randomly generated in the unit cube in {dim}-dimensions.\n \
Those with the sum of coordinates >= {dim}/2 are labeled 1, \n those below are \