[
https://issues.apache.org/jira/browse/FLINK-1745?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14956781#comment-14956781
]
ASF GitHub Bot commented on FLINK-1745:
---------------------------------------
Github user danielblazevski commented on a diff in the pull request:
https://github.com/apache/flink/pull/1220#discussion_r41984424
--- Diff:
flink-staging/flink-ml/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/LICENSE-2.0
+ *
+ * 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
+
+/**
+ * n-dimensional 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/CSCI1321-F11/Code/src/util/Quadtree.scala
+ *
+ * Many additional methods were added to the class both for
+ * efficient KNN queries and generalizing to n-dim.
+ *
+ * @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 --
@chiwanpark @tillrohrmann , sorry for the delay, but I had a look again at
making the changes Till suggested. In the case of multiplying a vector by a
double for the line of code ` + Childrennodes = Childrennodes :+ new
Node(cPart(i), L.map(x => x/2.0), null) ` Chiwan posted a workable solution.
On line 167 of QuadTree, there is also the statement:
``` scala
val root = new Node( (minVec, maxVec).zipped.map(_ + _).map(x=>0.5*x),
(maxVec, minVec).zipped.map(_ - _),null)
```
In pseudo-code, all that is needed is `(0.5*(MinVec + MaxVec), MaxVec -
MinVec)` Addition and multiplication by a double are not permitted for Vectors
at this moment. I could try to implement what Chiwan did for multiplying a
double, but that can get ugly and a lot of lines of code. I'm not sure if we
should try to add these operations into the Vector class with this PR, or apply
Chiwan's idea to this case as well.
> Add exact k-nearest-neighbours algorithm to machine learning library
> --------------------------------------------------------------------
>
> Key: FLINK-1745
> URL: https://issues.apache.org/jira/browse/FLINK-1745
> Project: Flink
> Issue Type: New Feature
> Components: Machine Learning Library
> Reporter: Till Rohrmann
> Assignee: Daniel Blazevski
> Labels: ML, Starter
>
> Even though the k-nearest-neighbours (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 (H-BNLJ, H-BRJ) algorithm as
> proposed in [2].
> Could be a starter task.
> Resources:
> [1] [http://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm]
> [2] [https://www.cs.utah.edu/~lifeifei/papers/mrknnj.pdf]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
