NumPy and SciPy for Data Mining and Data Analysis Including iPython, SciKits, and matplotlib
41 likes17,225 views
The document discusses the usage of NumPy and SciPy for data mining and analysis within the Python programming environment. It contrasts these tools with R, outlining their functionalities, installation, and features, as well as providing examples and demos of data analysis. Additionally, it highlights the advantages and disadvantages of using NumPy/SciPy versus R in data science contexts.
![ipython: a replacement CLI III
4 %pdb for spawning a pdb debug session.
5 Output cache saves the results of executing each line of
code. , , save the last three results.
6 Input cache allows re-execution of previously executed code
using a snapshot of the original input data in variable In. (to
rexecute lines 22 through 28, and line 34: exec In[22:29]
+ In[34].
7 Command history with(out) line numbers using %hist. Can
also toggle journaling with %logstart.
8 Execute system commands from within IPython as if you are
in the terminal by prefixing with !. Use !!, %sc, %sx to get
output into a Python variable (var = !ls -la). Use Python
variables when calling the shell.
Ryan R. Rosario
NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group](https://image.slidesharecdn.com/numpy-110911145719-phpapp01/85/NumPy-and-SciPy-for-Data-Mining-and-Data-Analysis-Including-iPython-SciKits-and-matplotlib-13-320.jpg)
NumPy and SciPy for Data Mining and Data Analysis Including iPython, SciKits, and matplotlib
- 1. Los Angeles R Users’ Group NumPy and SciPy for Data Mining and Data Analysis Including iPython, SciKits, and matplotlib Ryan R. Rosario March 29, 2011 Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 2. What is NumPy? NumPy is the standard package for scientific and numerical computing in Python: powerful N-dimensional array object. sophisticated (broadcasting) functions. tools for integrating C/C++ and Fortran code (R is similar) useful linear algebra, Fourier transform and random number capabilities. Additionally... data structures can be used as efficient multi-dimensional container of arbitrary data types. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 3. So, What is SciPy? Often (incorrectly) used as a synonym of NumPy. SciPy depends on NumPy’s data structures. OSS for mathematics, science, and engineering. provides efficient numerical routines for numerical integration and optimization. “NumPy provides the data structures, SciPy provides the application layer.” Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 4. Getting NumPy The ease of installation is highly dependent on OS and CPU. Official releases are recommended and available for 1 32 bit MacOS X Caveat: Must use Python Python, not Apple Python. 2 32 bit Windows Unofficial/third-party binary releases for 1 All Linux (use platform package system) 2 64 bit MacOS X (SciPy Superpack) Requires Apple Python 2.6.1 3 64-bit Windows (Christoph Gohlke, UC Irvine) Source Code Enthought is to SciPy/Numpy as Revolution is to R. Main download link is http://www.scipy.org/Download Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 5. Disclaimer Time is even more limited than usual, so the goal is to illustrate what NumPy is, and what NumPy, SciPy and its colleagues can achieve. Some goals: Brief NumPy tutorial via demo. Uses of NumPy and SciPy, including SciKits. Brief data analysis demo with NumPy, SciPy and matplotlib. Concluding remarks. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 6. NumPy Basics The main data structure is a homogeneous multidimensional array – a table of elements all of the same type, indexed by a tuple of positive integers. Some examples are vectors, matrices, images and spreadsheets. Why not just a use a list of lists? NumPy provides functions that operate on an entire array at once (like R), Python does not...really. NumPy is full of syntax and time is limited, so let’s jump straight into a demo... Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 7. NumPy Basics Demo Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 8. Keep in Mind: Broadcasting vs. Recycling In NumPy, if two arrays do not have the same length and we try to, say, add them, 1 is appended to the smaller array so that the dimensions match. In R, if two vectors do not have the same length, the values in the shorter vector are recycled and a warning is generated. Personally, recycling seems more practical. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 9. Keep in Mind: Indexing! R starts its indices at 1 and ranges are inclusive (i.e. seq(1,3) -> 1, 2, 3) NumPy (and Python) starts its indices at 0 and ranges are exclusive (i.e. arange(1,3) -> 1, 2). Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 10. Keep in Mind: Data Type Matters! R does a good job of using the correct data type. NumPy requires you to make decisions about data type. Using int64 for 1 is typically wasteful. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 11. ipython: a replacement CLI I IPython is the recommended CLI for NumPy/SciPy. I will use IPython in my demo. 1 an enhanced interactive Python shell. 2 an architecture for interactive parallel computing. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 12. ipython: a replacement CLI II Some advantages over the standard CLI: 1 tab completion for object attributes and filenames, auto parentheses and quotes for function calls. 2 simple object querying object name? prints details about an object. %pdoc (docstring), %pdef (function definition), %psource (full source code). 3 %run to run a code file, much like source in R. Differs from import because the code file is reread each time %run is called. Also provides special flags for profiling and debugging. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 13. ipython: a replacement CLI III 4 %pdb for spawning a pdb debug session. 5 Output cache saves the results of executing each line of code. , , save the last three results. 6 Input cache allows re-execution of previously executed code using a snapshot of the original input data in variable In. (to rexecute lines 22 through 28, and line 34: exec In[22:29] + In[34]. 7 Command history with(out) line numbers using %hist. Can also toggle journaling with %logstart. 8 Execute system commands from within IPython as if you are in the terminal by prefixing with !. Use !!, %sc, %sx to get output into a Python variable (var = !ls -la). Use Python variables when calling the shell. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 14. ipython: a replacement CLI IV 9 Easy access to the profiler, %prun. 10 Easy to integrate with matplotlib. 11 Custom profiles (options, modules to load, function definitions). Similar to R workspace. 12 Run other programs (that use pipes) and extract output (gnuplot). 13 Easily run doctests. 14 Lightweight version control. 15 Save/restore session, like in R (the authors specifically said “like R”). Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 15. matplotlib: SciPy’s Graphics System I 2D graphics library for Python that produces publication quality figures. Quality with respect to R is in the eye of the beholder. Graphics in R are much easier to construct, and require less code. Graphics in matplotlib may look nicer; require much more code. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 16. matplotlib: SciPy’s Graphics System II Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 17. matplotlib: SciPy’s Graphics System III Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 18. matplotlib: SciPy’s Graphics System IV Show matplotlib gallery. I will do a demo of matplotlib in the data analysis demo, if time. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 19. Quick Linear Regression Demo Demo here. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 20. SciKits I SciKits are modules or “plug-ins” for SciPy; they are too specialized to be included in SciPy itself. Below is a list of some relevant SciKits. 1 statsmodels: OLS, GLS, WLS, GLM, M-estimators for robust regression. 2 timeseries: time series analysis. 3 bootstrap: bootstrap error-estimation. 4 learn: machine learning and data mining Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 21. SciKits II 5 sparse: sparse matrices 6 cuda: Python interface to GPU powered libraries. 7 image: image processing 8 optimization: numerical optimization For more information, see http://scikits.appspot.com/scikits. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 22. Scientific Software Relying on SciPy I Many exciting software packages rely on SciPy Python Based Reinforcement Learning, Artificial Intelligence and Neural Network Library. Built around a neural networks kernel. Provides unsupervised learning and evolution. http://www.pybrain.org PyMC Pythonic Markov Chain Monte Carlo. MCMC, Gibbs sampling, diagnostics PyML: SVM, nearest neighbor, ridge regression, multiclass methods, model selection, filtering, CV, ROC. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 23. Scientific Software Relying on SciPy II Monte: gradient based learning machines, logistic regression, conditional random fields. MILK: SVM, k-NN, random forests, decision trees, feature selection, affinity propagation. Modular toolkit for Data Processing: signal processing, PCA, ICA, manifold learning, factor analysis, regression, neural networks, mixture models. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 24. Scientific Software Relying on SciPy III PyEvolve: Genetic algorithms Theano: Theano is a Python library that allows you to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently: GPUs, symbolic expressions, dynamic C code generation. Orange: Data mining, visualization and machine learning. Full GUI and Python scripting. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 25. Scientific Software Relying on SciPy IV SAGE: a hodgepodge of open-source software designed to be one analysis platform. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 26. Pros/Cons I What NumPy/SciPy does better: 1 Much larger community (SciPy + Python). 2 Better, and cleaner graphics...slightly. 3 Sits on atop a programming language rather than being one. 4 Dictionary data type is very useful (Python). 5 Adheres to a system of thought (“Pythonic”) that is more flexible. 6 Memory mapped files and extensive sparse matrix packages. 7 Easier to incorporate into a company/development workflow. 8 Opinion: Developers don’t have to dive into C as soon as they do with R. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 27. Pros/Cons II What R does better: 1 One word data.frame. 2 Installation is simple and universal. 3 Application is self-contained (NumPy consists of several moving parts). 4 Very simple, user friendly syntax. 5 Graphics and analysis are fairly quick to code. 6 Much easier data input. 7 Handles missing values better. Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 28. The SciPy community is very useful, prolific and friendly. Project has very high morale. Some places to get help: 1 http://www.numpy.org: extensive documentation, examples and cookbook. 2 http://www.scipy.org: extensive documentation, examples and cookbook. 3 NumPy and SciPy mailing lists are helpful. 4 StackOverflow 5 IRC channel #scipy 6 Frequent sprints and hackathons. 7 SciPyCon (Austin, TX) 8 SoCalPiggies and LA Python (first meeting April 1!) Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 29. Keep in Touch! My email: [email protected] My blog: http://www.bytemining.com Follow me on Twitter: @datajunkie Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group
- 30. References 1 NumPy http://www.numpy.org 2 SciPy http://www.scipy.org 3 SciKits http://scikits.appspot.com 4 iPython http://ipython.scipy.org/moin/ 5 matplotlib http://matplotlib.sourceforge.net/ Ryan R. Rosario NumPy/SciPy for Data Mining and Analysis Los Angeles R Users’ Group