def test_k_nearest_neighbor():
datasets = [
os.path.join('data', x) for x in os.listdir('data')
if os.path.splitext(x)[-1] == '.json'
]
aggregators = ['mean', 'mode', 'median']
distances = ['euclidean', 'manhattan', 'cosine']
for data_path in datasets:
# Load data and make sure its shape is correct
features, targets = load_json_data(data_path)
targets = targets[:, None] # expand dims
for d in distances:
for a in aggregators:
# make model and fit
knn = KNearestNeighbor(1, distance_measure=d, aggregator=a)
knn.fit(features, targets)
# predict and calculate accuracy
labels = knn.predict(features)
acc = accuracy(targets, labels)
# error if there's an issue
msg = 'Failure with dataset: {}. Settings: dist={}, agg={}.'.format(
data_path, d, a)
assert (acc == 1.0), msg
def test_perceptron():
features, targets = load_json_data('data/parallel_lines.json')
p = Perceptron(max_iterations=100)
p.fit(features, targets)
targets_hat = p.predict(features)
#your perceptron should fit this dataset perfectly
assert np.allclose(targets, targets_hat)
def test_transform_data():
features, targets = load_json_data('data/transform_me.json')
features_transform = transform_data(features)
p = Perceptron(max_iterations=100)
p.fit(features_transform, targets)
targets_hat = p.predict(features_transform)
#your perceptron should fit this dataset perfectly after transforming the data
assert np.allclose(targets, targets_hat)