2015 Data Science Summit @ dato Review
0 likes989 views
This document summarizes a data science summit attended by the author. It includes a brief overview of the author's travel itinerary to and from the event. The main body summarizes various sessions and topics discussed, including machine learning platforms and tools like Dato, IBM System ML, Apache Flink, PredictionIO, and DeepLearning4J. Session topics focused on scalable data processing, stream and batch processing, graph processing, and machine learning algorithms. The document provides links to several of the platforms and tools discussed.
2015 Data Science Summit @ dato Review
- 1. 2015 Data Science Summit @ dato
- 2. • 4:00am Got up and drove to LAX • 7:30am arrived SFO •8:20am ~ 6:30pm Data science summit • 9:00pm ~ 12:10am Flight delayed in SFO (-___-|||) • 1:30am Arrived LAX (still crowded!) A 22 Hours Trip
- 3. What are people talking about?
- 4. • Platform & system • Machine learning • Graph • Visualization • Scale • Spark • … Session Topics
- 5. Dato Data science stack https://dato.com/products/
- 6. • SArray SFrame and SGraph: Scalable External Memory Data Frame and Graph Structures for Machine Learning • GraphLab Create™ Translator https://dato.com/learn/translator/ GraphLab Create (ver. 1.0) vs Pandas (ver. 0.15.0) vs R (ver. 3.1.1) Dato SArray SFrame And SGraph
- 7. BDAS, the Berkeley Data Analytics Stack https://amplab.cs.berkeley.edu/software/
- 8. • IBM System ML, the machine learning platform developed at the company's facility in Almaden, is to be contributed to Apache Spark. IBM System ML http://www.theinquirer.net/inquirer/news/2413132/ibm-donates-machine-learning-tech-to-apache-spark- open-source-community
- 9. • Apache Flink is an open source platform for scalable batch and stream data processing. • Focused on large-scale data analytics • Unified real-time stream and batch processing • Expressive and rich APIs in Java / Scala (+Python) • Robust and fast execution backed • Some similar system storm? spark? … Apache Flink https://flink.apache.org/index.html
- 10. Python Data Science stack https://speakerdeck.com/jakevdp/the-state-of-the-stack-scipy-2015-keynote
- 11. Ibis on Impala: Python at Scale for Data Science http://blog.cloudera.com/blog/2015/07/ibis-on-impala-python-at-scale-for-data-science/ http://www.ibis-project.org/
- 12. • PredictionIO is an open source machine learning framework for developers and data scientists. It supports event collection, deployment of algorithms, evaluation, querying predictive results via REST APIs. • Built on Apache Spark, HBase and Spray. PredictionIO https://github.com/PredictionIO/PredictionIO/
- 13. • Splash is a general framework for parallelizing stochastic learning algorithms on multi-node clusters. • Splash is built on Scala and Apache Spark Splash http://zhangyuc.github.io/splash/
- 14. • Discriminative, low-dimensional, large-scale featurization – the easy way • A new way to process categorical features. It is a scaling way. Learning with Counts
- 15. Learning with Counts (Cont.)
- 16. •Three small wishes of data science. •Is more data always better? •Are more variables always better? •Can’t we keep at least some of our training performance? Statistics in the age of data science, issues you can not ignore
- 20. Other Interesting Sessions (Didn’t attend) • Applying data science to sales pipelines — for fun and profit http://es.slideshare.net/dato-inc/applying- data-science-to-sales-pipelines-for-fun-and-profit-51025224 • What we've learned from over 1MM machine learning models • DeepLearning4J: Open Source Neural Net Platform • Scalable on Hadoop, Spark and Akka + AWS et al • http://deeplearning4j.org/ • Five key learnings from building a large scale anomaly detection system http://www.slideshare.net/dato- inc/five-key-learnings-from-building-a-large-scale-anomaly-detection-system
- 22. Thanks!
Editor's Notes
- #11: Bokeh http://bokeh.pydata.org/en/latest/ Bokeh is a Python interactive visualization library that targets modern web browsers for presentation. dask https://github.com/ContinuumIO/dask Dask enables parallel computing through task scheduling and blocked algorithms.