![$
spark-‐shell
Welcome
to
____
__
/
__/__
___
_____/
/__
_
/
_
/
_
`/
__/
'_/
/___/
.__/_,_/_/
/_/_
version
1.2.0-‐SNAPSHOT
/_/
Using
Scala
version
2.10.4
(Java
HotSpot(TM)
64-‐Bit
Server
VM,
Java
1.7.0_67)
Type
in
expressions
to
have
them
evaluated.
Type
:help
for
more
information.
15/04/15
11:31:28
INFO
SparkILoop:
Created
spark
context..
Spark
context
available
as
sc.
scala>
val
a
=
sc.parallelize(1
to
100)
a:
org.apache.spark.rdd.RDD[Int]
=
ParallelCollectionRDD[0]
at
parallelize
at
<console>:12
scala>
a.collect.foreach(x
=>
println(x))
1
2
3
4](https://image.slidesharecdn.com/sparkmeetup-150420031004-conversion-gate02/85/Sparkly-Notebook-Interactive-Analysis-and-Visualization-with-Spark-9-320.jpg)
Sparkly Notebook: Interactive Analysis and Visualization with Spark
- 1. SPARKLY NOTEBOOK: INTERACTIVE ANALYSIS AND VISUALIZATION WITH SPARK FELIX CHEUNG APRIL 2015 HTTP://WWW.MEETUP.COM/SEATTLE-SPARK-MEETUP/EVENTS/208711962/
- 2. SETUP • Spark on CDH cluster • Vagrant - 2-nodes - custom provisioning
- 3. AGENDA • IPython + PySpark cluster • Zeppelin • Spark’s Streaming k-means • Lightning
- 5. SPARK - 10 SEC INTRODUCTION • Spark • Spark SQL + Data Frame + data source • Spark Streaming • MLlib • GraphX
- 6. It’s a lot of time looking at data..
- 8. Set of REPL related to Spark…
- 9. $ spark-‐shell Welcome to ____ __ / __/__ ___ _____/ /__ _ / _ / _ `/ __/ '_/ /___/ .__/_,_/_/ /_/_ version 1.2.0-‐SNAPSHOT /_/ Using Scala version 2.10.4 (Java HotSpot(TM) 64-‐Bit Server VM, Java 1.7.0_67) Type in expressions to have them evaluated. Type :help for more information. 15/04/15 11:31:28 INFO SparkILoop: Created spark context.. Spark context available as sc. scala> val a = sc.parallelize(1 to 100) a: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at parallelize at <console>:12 scala> a.collect.foreach(x => println(x)) 1 2 3 4
- 10. GOOD • See results instantly
- 11. NOT SO GOOD • Ok as an IDE • No Save / Repeat • No visualization
- 12. NOTEBOOK
- 14. Jupyter IPython will continue to exist as a Python kernel for Jupyter, but the notebook and other language-agnostic parts of IPython will move to new projects under the Jupyter name. IPython 3.0 will be the last monolithic release of IPython. ! “IPython” http://ipython.org/ • interactive shell • browser-based notebook • 'Kernel' • great support for visualization library (eg. matplotlib) • built on pyzmq, tornado IPYTHON/JUPYTER
- 15. IPYTHON NOTEBOOK NOTEBOOK == BROWSER-BASED REPL IPython Notebook is a web-based interactive computational environment for creating IPython notebooks. An IPython notebook is a JSON document containing an ordered list of input/output cells which can contain code, text, mathematics, plots and rich media.
- 16. MATPLOTLIB matplotlib tries to make easy things easy and hard things possible. You can generate plots, histograms, power spectra, bar charts, errorcharts, scatterplots, etc, with just a few lines of code, with familiar MATLAB APIs. plt.barh(y_pos, performance, xerr=error, align='center', alpha=0.4) plt.yticks(y_pos, people) plt.xlabel('Performance') plt.title('How fast do you want to go today?') plt.show()
- 17. PYSPARK • Spark on Python, this serves as the Kernel, integrating with IPython • Each notebook spins up a new instance of the Kernel (ie. PySpark running as the Spark Driver, in different deploy mode Spark/PySpark supports)
- 18. (All notebook examples are a subset of those in the Meetup reconstructed here)
- 19. Markdown
- 20. Spark in Python
- 23. WORD2VEC EXAMPLE Word2Vec computes distributed vector representation of words. Distributed vector representation is showed to be useful in many natural language processing applications such as named entity recognition, disambiguation, parsing, tagging and machine translation. https://code.google.com/p/word2vec/ Spark MLlib implements the Skip-gram approach. With Skip-gram we want to predict a window of words given a single word.
- 24. WORD2VEC DATASET Wikipedia dump http://mattmahoney.net/dc/ textdata grep -‐o -‐E 'w+(W+w+){0,15}' text8 > text8_lines then randomly sampled to ~200k lines
- 27. matplotlib: http://matplotlib.org Seaborn: http://stanford.edu/~mwaskom/software/seaborn/ Bokeh: http://bokeh.pydata.org/en/latest/ MORE VISUALIZATIONS Seaborn Bokeh matplotlib
- 28. SETUP To setup IPython • Python 2.7.9 (separate from CentOS default 2.6.6), on all nodes • matplotlib, on the host running IPython To run IPython with the PySpark Kernel, set these in the environment (Please check out my handy script on github) ! ! ! PYSPARK_PYTHON command to run python, eg. “python2.7” PYSPARK_DRIVER_PYTHON command to run ipython PYSPARK_DRIVER_PYTHON_OPTS “notebook —profile” PYSPARK_SUBMIT_ARGS pyspark commandline, eg. --master --deploy_mode YARN_CONF_DIR if YARN mode LD_LIBRARY_PATH for matplotlib
- 29. IPYTHON/JUPYTER KERNELS • IPython • IGo • Bash • IR • IHaskell • IMatlab • ICSharp • IScala • IRuby • IJulia .. and more https://github.com/ipython/ipython/wiki/IPython-kernels-for-other- languages
- 30. ZEPPELIN
- 31. Apache Zeppelin (incubating) is interactive data analytics environment for distributed data processing system. It provides beautiful interactive web-based interface, data visualization, collaborative work environment and many other nice features to make your data analytics more fun and enjoyable. Zeppelin has been incubating since Dec 2014. https://zeppelin.incubator.apache.org/
- 33. shell script & calling library package Load and process data with Spark
- 34. SQL query powered by Spark SQL - progress & parameterization via dynamic form
- 35. Python & data passing across languages (interpreters)
- 36. ZEPPELIN ARCHITECTURE Realtime collaboration - enabled by websocket communications Frontend: AngularJS Backend server: Java Interpreters: Java Visualization: NVD3
- 37. INTERPRETERS • Spark group • Spark (Scala) • PySpark • Spark SQL • Dependency • Markdownjs • Shell • Hive • Coming: jdbc, Tajo, etc.
- 38. CLUSTERING • Clustering tries to find natural groupings in data. It puts objects into groups in which those within a group are more similar to each other than to those in other groups. • Unsupervised learning
- 39. K-MEANS • First, given an initial set of k cluster centers, we find which cluster each data point is closest to • Then, we compute the average of each of the new clusters and use the result to update our cluster centers
- 41. K-MEANS|| IN MLLIB • a parallelized variant of the k-means++ http://theory.stanford.edu/~sergei/papers/vldb12-kmpar.pdf Parameters: • k is the number of desired clusters. • maxIterations is the maximum number of iterations to run. • initializationMode specifies either random initialization or initialization via k-means||. • runs is the number of times to run the k-means algorithm (k-means is not guaranteed to find a globally optimal solution, and when run multiple times on a given dataset, the algorithm returns the best clustering result). • initializationSteps determines the number of steps in the k-means|| algorithm. • epsilon determines the distance threshold within which we consider k- means to have converged.
- 42. CASE STUDY: K-MEANS - ZEPPELIN
- 43. Details on github at: http://bit.ly/1JWOPh8 ANOMALY DETECTION WITH K-MEANS Using Spark DataFrame, csv data source, to process KDDCup’99 data Scoring with different k values
- 45. Realtime updates
- 46. Dashboard
- 47. Spark-notebook: https://github.com/andypetrella/spark-notebook ISpark: https://github.com/tribbloid/ISpark Spark Kernel: https://github.com/ibm-et/spark-kernel Jove: https://github.com/jove-sh/jove-notebook Beaker: https://github.com/twosigma/beaker-notebook OTHER NOTEBOOKS • Spark-notebook • ISpark • Spark Kernel • Jove Notebook • Beaker • Databricks Cloud notebook
- 49. WHY STREAMING? • Train - model - predict works well on static data • What if data is • Coming in streams • Changing over time?
- 50. STREAMING K-MEANS DESIGN • Proposed by Dr Jeremy Freeman (here)
- 51. STREAMING K-MEANS • key concept: forgetfulness • balances the relative importance of new data versus past history • half-life • time it takes before past data contributes to only one half of the current model
- 52. STREAMING K-MEANS • time unit • batches (which have a fixed duration in time), or points • eliminate dying clusters
- 53. VISUALIZING STREAMING K-MEANS - LIGHTNING
- 54. LIGHTNING
- 55. • Lightning - data visualization server http://lightning-viz.org • provides API-based access to reproducible, web- based, interactive visualizations. It includes a core set of visualization types, but is built for extendability and customization. Lightning supports modern libraries like d3.js and three.js, and is designed for interactivity over large data sets and continuously updating data streams. VISUALIZING STREAMING K- MEANS ON IPYTHON + LIGHTNING
- 56. RUNNING LIGHTNING • API: node.js, Python, Scala • Extension support for custom chart (eg. d3.js) • Requirements: • Postgres recommended (SQLlite ok) • node.js (npm , gulp)
- 57. The Freeman Lab at Janelia Research Campus uses Lightning to visualize large-scale neural recordings from zebrafish, in collaboration with the Ahrens Lab
- 58. SPARK STREAMING K-MEANS DEMO Environment • requires: numpy, scipy, scikit-learn • IPython/Python requires: lightning-python package Demo consists of 3 parts: https://github.com/felixcheung/spark-ml-streaming • Python driver script, data generator • Scala job - Spark Streaming & Streaming k-means • IPython notebook to process result, visualize with Lightning Originally this was part of the Python driver script - it has been modified for this talk to run within IPython
- 61. CHALLENGES • Package management • Version/build conflicts!
- 62. YOU CAN RUN THIS TOO! • Notebooks available at http://bit.ly/1JWOPh8 • Everything is heavily scripted and automated Vagrant config for local, virtual environment available at http://bit.ly/1DB3OLw