An Incomplete Data Tools Landscape for Hackers in 2015
24 likes8,143 views
Wes McKinney gives an overview of the current data analysis tools landscape in Python and R. He discusses essential Python packages like NumPy, pandas, and scikit-learn. For R, he covers packages in the "Hadley stack" like dplyr and ggplot2. IPython/Jupyter notebooks are also mentioned as a platform for interactive data analysis across languages. The talk aims to highlight trends, opportunities, and challenges in the open source data science tool ecosystem.
![1
©
Cloudera,
Inc.
All
rights
reserved.
A
[Incomplete]
Data
Tools
Landscape
[for
Hackers]
in
2015
Wes
McKinney
@wesmckinn
Data^3
MeeMng
—
Minneapolis,
MN](https://image.slidesharecdn.com/slides-150307115108-conversion-gate01/85/An-Incomplete-Data-Tools-Landscape-for-Hackers-in-2015-1-320.jpg)
![2
©
Cloudera,
Inc.
All
rights
reserved.
This
talk
• A
parMal
look
at
different
languages
and
tools
• LimiMng
scope
to
either:
• Permissively
licensed
open
source
soSware,
e.g.
Apache-‐licensed
(OSS)
• Non-‐dual-‐licensed
copyleS
OSS
(e.g.
GPL)
• i.e.
“do
you
[the
community]
have
any
incenMve
to
create
patches?”
• Some
trends
(that
I
see,
anyway)
• Challenges
and
opportuniMes](https://image.slidesharecdn.com/slides-150307115108-conversion-gate01/85/An-Incomplete-Data-Tools-Landscape-for-Hackers-in-2015-2-320.jpg)
More Related Content
What's hot (20)
Viewers also liked (16)
Similar to An Incomplete Data Tools Landscape for Hackers in 2015 (20)
More from Wes McKinney (18)
Recently uploaded (20)
An Incomplete Data Tools Landscape for Hackers in 2015
- 1. 1 © Cloudera, Inc. All rights reserved. A [Incomplete] Data Tools Landscape [for Hackers] in 2015 Wes McKinney @wesmckinn Data^3 MeeMng — Minneapolis, MN
- 2. 2 © Cloudera, Inc. All rights reserved. This talk • A parMal look at different languages and tools • LimiMng scope to either: • Permissively licensed open source soSware, e.g. Apache-‐licensed (OSS) • Non-‐dual-‐licensed copyleS OSS (e.g. GPL) • i.e. “do you [the community] have any incenMve to create patches?” • Some trends (that I see, anyway) • Challenges and opportuniMes
- 3. 3 © Cloudera, Inc. All rights reserved. Who am I? • Python data firestarter • Financial analyMcs in R / Python starMng 2007 • pandas project born of frustraMon in 2008 • 2010-‐2012 • Hiatus from gainful employment • Make pandas ready for primeMme • Write "Python for Data Analysis"
- 4. 4 © Cloudera, Inc. All rights reserved. Who am I? (cont’d) • 2013-‐2014: Co-‐founder/CEO of DataPad (analyMcs startup, with early pandas collaborator Chang She) • Late 2014: DataPad team joins Cloudera • Now: backend systems and all-‐things-‐Python @ Cloudera
- 5. 5 © Cloudera, Inc. All rights reserved. SQL: SMll a lingua franca • “SQL: the Fortran of AnalyMcs” • OSen a concise, declaraMve way to express data transforms, analyMcs, etc. • RelaMvely easy to parse, analyze • SQL recently has seen resurgence with focus on interacMve-‐speed SQL engines, especially on top of HDFS/Hadoop • Relevant and impaclul features (e.g. JSON support) sMll arriving in established RDBMS like PostgreSQL
- 6. 6 © Cloudera, Inc. All rights reserved. Historical Python Context • ScienMfic / HPC compuMng focus in 1990s, 2000s • Python web community developed in parallel, matured faster! • NumPy became community standard in 2005, born from Numeric + Numarray • Pyrex, later Cython, easier C / C++ wrapping • f2py: easy Fortran wrapping • Anaconda distribuMon • Finally solving Python deployment for all
- 7. 7 © Cloudera, Inc. All rights reserved. EssenMal Python stack • NumPy: low-‐level array processing • SciPy: essenMal computaMonal algos • pandas: data wrangling • scikit-‐learn: machine learning • matplotlib (+ add-‐ons, like seaborn): visualizaMon • numba: numeric hotspot LLVM compiler • Domain-‐specific toolkits: nltk, scikit-‐image, statsmodels, Theano, PyCUDA/ PyOpenCL and many others
- 8. 8 © Cloudera, Inc. All rights reserved. pandas • A Pythonic take on the classic R “data frame” data structure • CriMcal piece to make the Python stack useful in everyday work • Added axis metadata / labeling for represenMng mulMdimensional data • Focus on easy data wrangling, IO, ploung, and basic analyMcs
- 9. 9 © Cloudera, Inc. All rights reserved. Jeff Reback’s “pandas as PyData middleware” diagram
- 10. 10 © Cloudera, Inc. All rights reserved. Newer / Up-‐and-‐coming Python projects • Bokeh: interacMve / reacMve visualizaMon for the web • Blaze: uniform data expression API • Odo: easy data migraMon
- 11. 11 © Cloudera, Inc. All rights reserved. R Project • Trusted base of staMsMcs libraries • Latest and greatest stats research oSen hits R first • RStudio • The "Hadley stack” • VisualizaMon: ggplot2 (staMc) and ggvis (interacMve) • Data Wrangling: dplyr • legacy: plyr / reshape2
- 12. 12 © Cloudera, Inc. All rights reserved. dplyr • Started late 2012 by Hadley Wickham, supported by RStudio • Composable / chainable analyMcs and data wrangling expressions • In-‐memory and SQL backends • Has avracted folks back to R from Python in a lot of cases
- 13. 13 © Cloudera, Inc. All rights reserved. Some other great R stuff • shiny: interacMve web apps in R • Rcpp • data.table • xts
- 14. 14 © Cloudera, Inc. All rights reserved. IPython • IPython started out as a bever interacMve Python • Grew to include web-‐based computaMonal notebook, GUI console, and other components • (Google even integrated into Google Drive!) • IPython Notebook architecture enabled “kernel” processes to be wriven in nearly any language (even bash!) • How to build community beyond Python?
- 15. 15 © Cloudera, Inc. All rights reserved. Enter Jupyter • hvp://jupyter.org • Breaking out notebook machinery into a standalone non-‐Python-‐specific project • Enable project components to evolve at own pace, without large monolithic releases • JupyterHub: upcoming mulM-‐user notebook server
- 16. 16 © Cloudera, Inc. All rights reserved. A few words about Hadoop + Big Data
- 17. 17 © Cloudera, Inc. All rights reserved. Apache Spark • Originated from Berkeley AMPLab • General purpose distributed memory-‐centric data processing framework • Official APIs: Scala, Java, Python Source: databricks.com
- 18. 18 © Cloudera, Inc. All rights reserved. Spark 1.3: DataFrames! • R/pandas-‐inspired API for tabular data manipulaMon in Scala, Python, etc. • Logical operaMon graphs rewriven internally in more efficient form • Good interop with Spark SQL • Some interoperability with pandas • Will help close the semanMc gap between Spark and R/Python
- 19. 19 © Cloudera, Inc. All rights reserved. Some problems in need of solving • A Shiny-‐like quick-‐and-‐dirty data app development framework for Python • IPython/Jupyter notebook collaboraMon • A community-‐standard, Apache-‐licensed C/C++ data frame library with best-‐in-‐ class performance • Ubiquitous support for emerging analyMcal on-‐disk storage standards like Parquet
- 20. 20 © Cloudera, Inc. All rights reserved. Other interesMng stuff to look at • Torch7 / LuaJIT: high performance ML / deep learning on GPUs • Facebook AI group open sourced several ML modules • Apache Flink • Up-‐and-‐coming Scala-‐based data processing framework • Some overlap with Spark use cases
- 21. 21 © Cloudera, Inc. All rights reserved. Some other interesMng industry trends • MicrosoS • Acquired RevoluMon AnalyMcs, leading commercial R vendor • Launched Azure ML: R, Python, and more on Azure cloud • Dato (ya GraphLab) • faster, more scalable machine learning, with Python interface (Paid commercial product, free for non-‐commercial/academic use) • Largest-‐ever VC investment in a data tools company beung big on Python • Databricks • Offering cloud Spark-‐notebook-‐as-‐a-‐service
- 22. 22 © Cloudera, Inc. All rights reserved. Thank you @wesmckinn