A Python wrapper for MADlib - an open source library for scalable in-database machine learning algorithms

PyMADlib currently has wrappers for the following algorithms in MADlib

1. Linear regression
2. Logistic Regression
3. SVM (regression & classification)
4. K-Means
5. LDA

Refer MADlib User Docs for MADlib's user documentation.

1. You'll need the python extension psycopg2 to use PyMADlib.
2. If you have matplotlib installed, you'll see Matplotlib visualizations for Linear Regression demo.
3. If you have installed networkx, you'll see a visualization of the k-means demo
4. PyROC is included in the source of this distribution with permission from its developer. You'll see a visualization of the ROC curves for Logistic Regression.

To configure your DB Connection parameters You should create a file in your home directory

~/.pymadlib.config 

that should look like so :

[db_connection]  
user = gpadmin  
password = XXXXX  
hostname = 127.0.0.1 (or the IP of your DB server)  
port = 5432 (the port# of your DB)  
database = vatsandb (the database you wish to connect to)  

PyMADlib depends on psycopg2 and Pandas. It is easiest to work with PyMADlib if you have Anaconda Python.

• Download & install [Anaconda-1.9.0-MacOSX-x86_64.pkg] (http://repo.continuum.io/archive/Anaconda-1.9.0-MacOSX-x86_64.pkg)

• Open a terminal and check if you have Anaconda Python & the package manager conda

vatsan-mac$ which python
/Users/vatsan/anaconda/bin/python
vatsan-mac$ which conda
/Users/vatsan/anaconda/bin/conda 

• If you haven't installed PostgreSQL on your Mac already, you'll have to download & install PostGreSQL for Mac. This is so that we get some required libraries to compile the SQL Engine: psycopg2. The easiest way to install PostGreSQL on Mac is via http://postgresapp.com/. Once you've downloaded and installed PostGreSQL on Mac, it should typically be found under /Library/PostgreSQL

vatsan-mac$ ls /Library/PostgreSQL/9.2/
Library include pg_env.sh uninstall-postgresql.app
bin installer scripts
data lib share
doc pgAdmin3.app stackbuilder.app

I don't think the version of the PostGreSQL matters (9.1 or above is fine).

• You may need to create some symlinks to libpq & libssl so that psycopg2 is able to find it:

vatsan-mac$ sudo ln -s /Users/vatsan/anaconda/lib/libssl.1.0.0.dylib /usr/lib
vatsan-mac$ sudo ln -s /Users/vatsan/anaconda/lib/libcrypto.1.0.0.dylib /usr/lib

• Install Psycopg2

vatsan-mac$ conda install distribute
vatsan-mac$ pip install psycopg2

• Now we're ready to test if the installations of the required libraries were successful.

vatsan-mac$ python -c 'import psycopg2'

If the above command did not error out, then installation was successful.

• You may install PyMADlib by downloading the source (from PyPI) and then run the following

sudo python setup.py build
sudo python setup.py install

• If you use easy_install or pip, simply run :

sudo easy_install pymadlib

Visit PyMADlib Tutorial for a tutorial on using PyMADlib Also visit PyMADlib IPython NB to download the IPython NB tutorial

You may run the demo from the extracted directory of pymadlib like so :

python example.py

If you installed PyMADlib using instructions in the previous section, then simply run

python -c 'from pymadlib.example import runDemos; runDemos()'

Remember to close the Matplotlib windows that pop-up to continue with the rest of the demo.

PyMADlib packages publicly available datasets from the UCI machine learning repository and other sources.

1. Wine quality dataset from UCI Machine Learning repository
2. Auto MPG dataset from UCI ML repository from UCI Machine Learning repository
3. Wine quality dataset from UCI Machine Learning repository
4. Obama-Romney second presidential debate (2012) transcripts

[email protected]