[ https://issues.apache.org/jira/browse/FLINK1745?page=com.atlassian.jira.plugin.system.issuetabpanels:commenttabpanel&focusedCommentId=14956937#comment14956937
]
ASF GitHub Bot commented on FLINK1745:

Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/1220#discussion_r41992363
 Diff: flinkstaging/flinkml/src/main/scala/org/apache/flink/ml/nn/QuadTree.scala

@@ 0,0 +1,305 @@
+
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.flink.ml.nn.util
+
+import org.apache.flink.ml.math.Vector
+import org.apache.flink.ml.metrics.distances.DistanceMetric
+
+import scala.collection.mutable.ListBuffer
+import scala.collection.mutable.PriorityQueue
+
+/**
+ * ndimensional QuadTree data structure; partitions
+ * spatial data for faster queries (e.g. KNN query)
+ * The skeleton of the data structure was initially
+ * based off of the 2D Quadtree found here:
+ * http://www.cs.trinity.edu/~mlewis/CSCI1321F11/Code/src/util/Quadtree.scala
+ *
+ * Many additional methods were added to the class both for
+ * efficient KNN queries and generalizing to ndim.
+ *
+ * @param minVec
+ * @param maxVec
+ */
+class QuadTree(minVec:ListBuffer[Double], maxVec:ListBuffer[Double],distMetric:DistanceMetric){
+ var maxPerBox = 20
+
+ class Node(c:ListBuffer[Double],L:ListBuffer[Double], var children:ListBuffer[Node])
{
+
+ var objects = new ListBuffer[Vector]
+
+ /** for testing purposes only; used in QuadTreeSuite.scala
+ *
+ * @return
+ */
+ def getCenterLength(): (ListBuffer[Double], ListBuffer[Double]) = {
+ (c, L)
+ }
+
+ def contains(obj: Vector): Boolean = {
+ overlap(obj, 0.0)
+ }
+
+ /** Tests if obj is within a radius of the node
+ *
+ * @param obj
+ * @param radius
+ * @return
+ */
+ def overlap(obj: Vector, radius: Double): Boolean = {
+ var count = 0
+ for (i < 0 to obj.size  1) {
+ if (obj(i)  radius < c(i) + L(i) / 2 && obj(i) + radius > c(i)
 L(i) / 2) {
+ count += 1
+ }
+ }
+
+ if (count == obj.size) {
+ return true
+ } else {
+ return false
+ }
+ }
+
+ /** Tests if obj is near a node: minDist is defined so that every point in the box
+ * has distance to obj greater than minDist
+ * (minDist adopted from "Nearest Neighbors Queries" by N. Roussopoulos et al.)
+ *
+ * @param obj
+ * @param radius
+ * @return
+ */
+ def isNear(obj: Vector, radius: Double): Boolean = {
+ if (minDist(obj) < radius) {
+ true
+ } else {
+ false
+ }
+ }
+
+ def minDist(obj: Vector): Double = {
+ var minDist = 0.0
+ for (i < 0 to obj.size  1) {
+ if (obj(i) < c(i)  L(i) / 2) {
+ minDist += math.pow(obj(i)  c(i) + L(i) / 2, 2)
+ } else if (obj(i) > c(i) + L(i) / 2) {
+ minDist += math.pow(obj(i)  c(i)  L(i) / 2, 2)
+ }
+ }
+ return minDist
+ }
+
+ def whichChild(obj:Vector):Int = {
+
+ var count = 0
+ for (i < 0 to obj.size  1){
+ if (obj(i) > c(i)) {
+ count += Math.pow(2,i).toInt
+ }
+ }
+ count
+ }
+
+ def makeChildren() {
+ var cBuff = new ListBuffer[ListBuffer[Double]]
+ cBuff += c
+ var Childrennodes = new ListBuffer[Node]
+ val cPart = partitionBox(cBuff,L,L.length)
+ for (i < cPart.indices){
+ Childrennodes = Childrennodes :+ new Node(cPart(i), L.map(x => x/2.0), null)
+
+ }
 End diff 
Hi @danielblazevski, sorry for not being response either. We just had our first conference.
You're right that with the current `Vector` interface of FlinkML you cannot do much if
you want to apply algebraic operations. The recommended way to do it at the moment is to convert
Flink `Vectors` into `BreezeVectors` which give you all the operations you can think of. It
is very convenient to convert from and to `BreezeVector`. Simply add `import org.apache.flink.ml.math.Breeze._`
to your sources and then you can call `vector.asBreeze` to obtain a `BreezeVector` from a
`Vector` and `breezeVector.fromBreeze` to obtain a `Vector` from a `BreezeVector`. Then you
can simply perform all these algebraic operations without having to operate on the underlying
data structure.
One thing to note is that the conversion is not so costly, because we use the same data
representation as breeze for our vectors. Therefore, the object creation is rather cheap.
However, it is better to not convert in every step from and to breeze. It's better to convert
once to breeze in the beginning and then in the end convert back to Flink. You can take a
look at the `SVM` code to see how we used it there.
> Add exact knearestneighbours algorithm to machine learning library
> 
>
> Key: FLINK1745
> URL: https://issues.apache.org/jira/browse/FLINK1745
> Project: Flink
> Issue Type: New Feature
> Components: Machine Learning Library
> Reporter: Till Rohrmann
> Assignee: Daniel Blazevski
> Labels: ML, Starter
>
> Even though the knearestneighbours (kNN) [1,2] algorithm is quite trivial it is still
used as a mean to classify data and to do regression. This issue focuses on the implementation
of an exact kNN (HBNLJ, HBRJ) algorithm as proposed in [2].
> Could be a starter task.
> Resources:
> [1] [http://en.wikipedia.org/wiki/Knearest_neighbors_algorithm]
> [2] [https://www.cs.utah.edu/~lifeifei/papers/mrknnj.pdf]

This message was sent by Atlassian JIRA
(v6.3.4#6332)
