Computer – Page 2

TensorFlow For Poets – Retrain Inception behind a Proxy

I tried to retrain Google’s Inception as described here. I failed since I use a proxy what has not been considered when implementing the retrain.py script.

So what I did to solve it is to find the following line in the script:

filepath, _ = urllib.request.urlretrieve(DATA_URL, filepath, _progress)

… and add the following code above the line:

proxy = urllib.request.ProxyHandler({'http': r'http://user:password@proxy.domain.de:8080'})
auth = urllib.request.HTTPBasicAuthHandler()
opener = urllib.request.build_opener(proxy, auth, urllib.request.HTTPHandler)
urllib.request.install_opener(opener)

Now tensorflow can download the required files.

Install Python, Scikit and Tensorflow on Windows

Lately, I worked through Google‘s fantastic machine learning tutorial by Josh Gordon that you can find here. On a windows machine installing the required components is not easy and documented somewhat cryptically. As a consequence, I decided to write the necessary steps down so that you can step through the easily.

Download and install Python 3.5.3 from the official website. Tensorflow requires us to use version 3.5.x on windows so make sure that you don’t use 2.x or 3.6.x.
Install scikit-learn, tensorflow and pydotplus from the command line: pip install scikit-learn pydotplus tensorflow If you are using a proxy call pip with the proxy parameter (e.g. pip –proxy http://user:password@proxy.mydomain.com:8080 install scikit-learn pydotplus tensorflow)
Install GraphViz from the official website and add its binary folder to your PATH variable (e.g. C:\Program Files (x86)\Graphviz2.38\bin).
Download and install the compiled binaries for numpy 1.13 and scipy 0.19.0 from Christoph Gohlke‘s website. Higher versions might work as well. By downloading the pre-compiled binaries we avoid to install multiple compilers on our system. When downloading make sure to grab the right packages for Python 3.5 (…cp35…) and win64.
Install both packages using pip:
1. pip install C:\Users\admin\Downloads\numpy-1.13.0rc2+mkl-cp35-cp35m-win_amd64.whl
2. pip install C:\Users\admin\Downloads\scipy-0.19.0-cp35-cp35m-win_amd64.whl

Thats it, not you can use Scikit and Tensorflow on windows!

Are points of a triangle in clockwise order?

Doing some research on triangulation algorithms I found an interesting code snipped that I’d like to share with you. Its intention is to determine if the three points of a triangle are in (counter-)clockwise order.

double Math::crossProductZ(const Vector2 &amp;a, const Vector2 &amp;b) {
	return a.x * b.y - a.y * b.x;
}
 
// If orientation is positive, if a-&gt;b-&gt;c is counterclockwise.
double Math::orientation(const Vector2 &amp;a, const Vector2 &amp;b, const Vector2 &amp;c) {
        return Math::crossProductZ(a, b) +
               Math::crossProductZ(b, c) +
               Math::crossProductZ(c, a);
}

Vector2 is a simple class with x and y coordinates. The method crossProductZ calculates the z component of the cross product of both Vector2s while orientation sums them up. If orientation returns a positive result the points are sorted counterclockwise, otherwise clockwise.

This method is helpful if you write code for different game engines which require an opposite order of indices to set the orientation (the normal) of each triangle in a mesh.

Java client for the hbase REST API

Modern open source projects frequently lack documentation and so does hbase regarding its REST API. Even though REST is covered in the official hbase reference book the guidelines are not complete as I’ll show in the listings below. A second source that you might have come across is a set of three blog posts by Cloudera which is quite good and much more detailled than the official hbase guidelines – but it still does not fully cover then traps you might run into when implementing a java client.

I’d like to show three examples, two for a put and one for a get request. All further operations like delete or post should then be easy to implement once you know how to send requests to the REST server.

Continue reading “Java client for the hbase REST API”

Java / Spark – Printing RDDs from Spark Streaming

A common challenge when writing functional code in Spark is to simply output logs as we usually do it in ordinary applications where line for line is processed sequentially and a debug message could be printed at an arbitrary place using System.out.println(). Working with Spark’s RDDs is new to most of us and might be confusing at first since it is difficult to track what object type we are currently working with (is it a RDDStream, an RDD or a simple type like String or int) – but infact this tracking is important. If you make use of the print function that is offered by RDDs or RDDStreams you’ll just see more or less useless array like information in the console instead of the objects’ content.

What I commonly do is to resolve the RDD object down to a simple JavaRDD or JavaDStream<object> using e.g. flatmap(x -> x) to flatten streams or lists in a dataset or map() to break down complex objects to a simple data type (e.g. via map(person -> person.getAddress().getStreet())). On RDDs you can then do a foreach(System.out::println). On JavaDStream you can do the following:

public static void printRDD(final JavaRDD<String> s) {
  s.foreach(System.out::println);
}
 
public static void main(final String[] args) {
  SparkConf conf = new SparkConf().setMaster("local[*]").setAppName("testApp");
  try (JavaStreamingContext sc = new JavaStreamingContext(conf, new Duration(5000))) {
    JavaDStream<String> stream = sc.textFileStream("C:/testnumbers/");
    stream.foreachRDD(Main::printRDD);
    sc.start();
    sc.awaitTermination();
  }
}

Here, I simply break down a JavaDStream to an RDD using foreachRDD() and use System.out::println as method reference to print each RDDs content.

Java – Streams vs Collections vs Loops

Trying to figure out what method serves best in java to get the maximum timestamp out of a list I setup a small benchmark to evaluate the runtime of Java Streams, collections and an ordinary for loop. My setup can be seen in the following listing.

public static void main(final String[] args) {
 
  List list = new ArrayList();
  Random rnd = new Random();
  for (int i = 0; i < 10; i++) { list.add(rnd.nextInt(10000)); } long t1 = System.currentTimeMillis(); long largestNumber = 0; for (Integer t : list) { if (t > largestNumber) {
      largestNumber = t;
    }
  }
 
  long t2 = System.currentTimeMillis();
  System.out.println("Max (loop) is " + largestNumber + " in " + (t2 - t1) + " ms.");
 
  long t3 = System.currentTimeMillis();
  OptionalInt max = list.stream().mapToInt(Integer::intValue).max();
 
  long t4 = System.currentTimeMillis();
  if (max.isPresent()) {
    System.out.println("Max (Stream) is " + max.getAsInt() + " in " + (t4 - t3) + " ms.");
  }
 
  long t5 = System.currentTimeMillis();
  int collMax = Collections.max(list);
  long t6 = System.currentTimeMillis();
 
  System.out.println("Max (Collections) is " + collMax + " in " + (t6 - t5) + " ms.");
}

Here are the results:

Samples	Steams	Collections	Loop
10.000	68	3	3
50.000	70	5	4
100.000	71	6	5
500.000	76	16	10
1.000.000	108	18	12
1.500.000	89	20	17
5.000.000	94	29	19
10.000.000	103	41	30

The expected linear runtime allows a comparision of these three ways. Although, collections and streams look way better than the loop runtime shows that it still seems to be the fastest way to determine a max value from values in a simple list.

Apache Ranger 0.5 build errors – How to fix them

I recently came accross some maven errors while building Apache Ranger 0.5 leading to a failure while building the HDFS Security Plugin and the Storm Security Plugin.

To get rid of the error just change the version of the maven-assembly-plugin in Ranger’s main pom.xml from

<version>2.2-beta-5</version>

<version>2.3</version>

…, add the following repository definitions to the main pom.xml

<repository>
<id>conjars</id>
<name>Concurrent Conjars repository</name>
<url>http://conjars.org/repo</url>
<layout>default</layout>
</repository>
<repository>
<id>clojars.org</id>
<url>http://clojars.org/repo</url>
</repository>

…, add the following dependency to the storm-agent/pom.xml:

<dependency>
<groupId>ring</groupId>
<artifactId>ring-jetty-adapter</artifactId>
<version>0.3.11</version>
</dependency>

Now you should be able to build ranger. If you are facing issues with a missing pom.xml for the dependency tomcat:common-el then add the following repository to the main pom.xml:

<repository>
<id>conjars</id>
<name>Concurrent Conjars repository</name>
<url>http://conjars.org/repo</url>
<layout>default</layout>
</repository>

The absence of the pom.xml on the default maven repository should only be a temporary issue but it took me some time to figure it out so that you can avoid it easily by addind a second repo that provides the missing file.

Wichtelwald – Contest entry for Devmania 2015

Last weekend we wrote a game for the over-night contest of the Devmania game jam in Mainz. Collect wood and snowballs at day time to be able to defend your hut at night against the evil goblins. The gameplay is a funny mixture of isometric collectable and first person tower defence.

Download it here.

HDFS – setgid and setuid still not possible in Hadoop 2.7.1

I faced the requirement to set the setgid and/or the setuid for directories in a customer’s HDFS so that I spent the last 10 minutes searching the internet for news on a possibility to do so. Long story short: results were bad. Last time I tried with Hadoop 2.2.0 without success, today I tried with 2.7.1 but there were no changes at all. So commands like the following are NOT possible:

hdfs dfs -chmod g+s /tmp
hdfs dfs -chmod u+s /tmp

As a matter of fact, files created in HDFS receive the group of the parent directory and hence act as if the setgid flag is set. The uid a newly created referes always to the user that created the file.

Update: When mounting an HDFS as NFS Gateway the behavior is different since the file system acts as if the setgid was NOT set. Which is strange since same file operations return different results depending on on which interface the operation is executed on. Furthermore, the NFS Gateway does not take secondary groups into account when evaluating permissions to write into the mounted HDFS. The only options to achive an allowance is a user mapping as described at the bottom of the page of the Hadoop Permissions Guide.

If you face the need to make use of the flags think about using a posix compliant alternative to HDFS like Spectrum Scale/GPFS.

Deploy a web application to an embedded Tomcat 8

I’ve found various tutorials on creating an embedded tomcat programatically and deploy a web application to it. Unfortunately, non of the tutorials was up to date using Tomcat 8 and actually non was showing how to deploy anything else but a servlet so I decided to write a short tutorial on how to create the server and deploy either a WAR file or a folder containing a non archived web application.

Create a simple Java project in ecplipse, enable maven and add the following dependencies:

<dependencies>
  <dependency>
    <groupId>org.apache.tomcat</groupId>
    <artifactId>tomcat-catalina</artifactId>
    <version>8.0.21</version>
  </dependency>
  <dependency>
    <groupId>org.apache.tomcat</groupId>
    <artifactId>tomcat-util</artifactId>
    <version>8.0.21</version>
  </dependency>
  <dependency>
    <groupId>org.apache.tomcat.embed</groupId>
    <artifactId>tomcat-embed-core</artifactId>
    <version>8.0.21</version>
  </dependency>
  <dependency>
    <groupId>org.apache.tomcat.embed</groupId>
    <artifactId>tomcat-embed-jasper</artifactId>
    <version>7.0.8</version>
  </dependency>
</dependencies>

Optionally, you can enable the maven-assembly-plugin to package your application as JAR at the end and have maven include all dependencies you specified. As build goal you’ll have to use assembly:single.

<plugin>
  <artifactId>maven-assembly-plugin</artifactId>
  <configuration>
    <archive>
      <manifest>
        <mainClass>de.jofre.embeddedtc.runtime.Main</mainClass>
      </manifest>
    </archive>
    <descriptorRefs>
      <descriptorRef>jar-with-dependencies</descriptorRef>
    </descriptorRefs>
  </configuration>
</plugin>

Now create the class de.jofre.embeddedtc.runtime.Main (The name is arbitrary) and add write the code according to the next listing:

package de.jofre.embedded.runtime;
 
import java.io.File;
import java.util.logging.Logger;
 
import org.apache.catalina.Context;
import org.apache.catalina.LifecycleException;
import org.apache.catalina.startup.Tomcat;
 
public class Main {
  private final static Logger LOGGER = Logger.getLogger(Main.class.getName());
  private final static String mWorkingDir = System.getProperty("java.io.tmpdir");
  private static Tomcat tomcat = null;
 
  public static void main(String[] args) {
 
    tomcat = new Tomcat();
    tomcat.setPort(8080);
    tomcat.setBaseDir(mWorkingDir);
    tomcat.getHost().setAppBase(mWorkingDir);
    tomcat.getHost().setAutoDeploy(true);
    tomcat.getHost().setDeployOnStartup(true);
 
    try {
      tomcat.start();
    } catch (LifecycleException e) {
      LOGGER.severe("Tomcat could not be started.");
      e.printStackTrace();
    }
    LOGGER.info("Tomcat started on " + tomcat.getHost());
 
    // Alternatively, you can specify a WAR file as last parameter in the following call e.g. "C:\\Users\\admin\\Desktop\\app.war"
    Context appContext = Main.getTomcat().addWebapp(Main.getTomcat().getHost(), "/app", "C:\\Users\\admin\\Desktop\\app\\");
    LOGGER.info("Deployed " + appContext.getBaseName() + " as " + appContext.getBaseName());
 
    tomcat.getServer().await();
  }
}

The last question is how the directory in C:\\Users\\admin\\Desktop\\app\\ respectively the C:\\Users\\admin\\Desktop\\app.war looks like. Well, it contains a simple HTML file…

<html><body>Test</body></html>

… and another folder called WEB-INF containing the web.xml with the following content:

<?xml version="1.0" encoding="ISO-8859-1"?>
 
<!DOCTYPE web-app 
    PUBLIC "-//Sun Microsystems, Inc.//DTD Web Application 2.3//EN" 
    "http://java.sun.com/dtd/web-app_2_3.dtd">
 
<web-app>
 
    <display-name>Test App</display-name>
    <description>A test app</description>
 
	<welcome-file-list>
			<welcome-file>index.html</welcome-file>
	</welcome-file-list>
 
    <session-config>
      <session-timeout>30</session-timeout>
    </session-config>
 
</web-app>

Now if you start the java application you call http://localhost:8080/app and the content of index.html should be displayed. Hope this helps you!