# Copyright (C) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License. See LICENSE in project root for information.
import sys
from pyspark import SparkContext
if sys.version >= '3':
basestring = str
[docs]class ConditionalBallTree(object):
def __init__(self, keys, values, labels, leafSize, java_obj=None):
"""
Create a conditional ball tree.
:param keys: 2D array representing the data, shape: n_points x n_features
:param values: 1D array
:param labels: 1D array
:param leafSize: int
"""
if java_obj is None:
self._jconditional_balltree = SparkContext._active_spark_context._jvm \
.com.microsoft.ml.spark.nn.ConditionalBallTree \
.apply(keys, values, labels, leafSize)
else:
self._jconditional_balltree = java_obj
[docs] def findMaximumInnerProducts(self, queryPoint, conditioner, k):
"""
Find the best match to the queryPoint given the conditioner and k from self.
:param queryPoint: array vector to use to query for NNs
:param conditioner: set of labels that will subset or condition the NN query
:param k: int representing the maximum number of neighbors to return
:return: array of tuples representing the index of the match and its distance
"""
return [(bm.index(), bm.distance())
for bm in self._jconditional_balltree.findMaximumInnerProducts(queryPoint, conditioner, k)]
[docs] def save(self, filename):
self._jconditional_balltree.save(filename)
[docs] @staticmethod
def load(filename):
java_obj = SparkContext._active_spark_context._jvm \
.com.microsoft.ml.spark.nn.ConditionalBallTree.load(filename)
return ConditionalBallTree(None, None, None, None, java_obj=java_obj)