def build_tree(data, centers, distances, valid_centers, valid_data):
node = Node()
if valid_centers.sum() == 1:
node.value = np.argmax(valid_centers)
return node
dim, cut, cost = best_cut(data, valid_data, centers,
valid_centers, distances)
node.feature = dim
node.value = cut
n = data.shape[0]
left_valid_data = np.zeros(n, dtype=bool)
right_valid_data = np.zeros(n, dtype=bool)
for i in range(n):
if valid_data[i]:
if data[i,dim] <= cut:
left_valid_data[i] = True
else:
right_valid_data[i] = True
k = centers.shape[0]
left_valid_centers = np.zeros(k, dtype=bool)
right_valid_centers = np.zeros(k, dtype=bool)
for i in range(k):
if valid_centers[i]:
if centers[i, dim] <= cut:
left_valid_centers[i] = True
else:
right_valid_centers[i] = True
node.left = build_tree(data, centers, distances,
left_valid_centers, left_valid_data)
node.right = build_tree(data, centers, distances,
right_valid_centers, right_valid_data)
return node
