Wednesday, February 26, 2014

Terms selection with chi-square

In Natural Language Processing, the identification the most relevant terms in a collection of documents is a common task. It can produce meaningful insights about the data and it can also be useful to improve classification performances and computational efficiency. A popular measure of relevance for terms is the χ2 statistic. To compute it we can convert the terms of our document collection and turn them into features of a vectorial model, then χ2 can be computed as follow:

Where f is a feature (a term in this case), t is a target variable that we, usually, want to predict, A is the number of times that f and t cooccur, B is the number of times that f occurs without t, C is the number of times that t occurs without f, D is the number of times neither t or f occur and N is the number of observations.

Let's see how χ2 can be used through a simple example. We load some posts from 4 different newsgroups categories using the sklearn interface:
from sklearn.datasets import fetch_20newsgroups
 # newsgroups categories
categories = ['alt.atheism','talk.religion.misc',

posts = fetch_20newsgroups(subset='train', categories=categories,
                           shuffle=True, random_state=42,
From the posts loaded, we build a linear model using all the terms in the document collection but the stop words:
from sklearn.feature_extraction.text import CountVectorizer
vectorizer = CountVectorizer(lowercase=True,stop_words='english')
X = vectorizer.fit_transform(
Now, X is a document-term matrix where the element Xi,j is the frequency of the term j in the document i. Then, the features are given by the columns of X and we want to compute χ2 between the categories of interest and each feature in order to figure out what are the most relevant terms. This can be done as follows
from sklearn.feature_selection import chi2
# compute chi2 for each feature
chi2score = chi2(X,[0]
To have a visual insight, we can plot a bar chart where each bar shows the χ2 value computed above:
from pylab import barh,plot,yticks,show,grid,xlabel,figure
wscores = zip(vectorizer.get_feature_names(),chi2score)
wchi2 = sorted(wscores,key=lambda x:x[1]) 
topchi2 = zip(*wchi2[-25:])
x = range(len(topchi2[1]))
labels = topchi2[0]

We can observe that the terms with a high χ2 can be considered relevant for the newsgroup categories we are analyzing. For example, the terms space, nasa and launch can be considered relevant for the group The terms god, jesus and atheism can be considered relevant for the groups alt.atheism and talk.religion.misc. And, the terms image, graphics and jpeg can be considered relevant in the category

Tuesday, January 14, 2014

Review: Fundamentals of Data Analytics in Python

I massively use Python for data analysis and when I was offered to review the video tutorial with the title “Fundamentals of Data Analytics in Python LiveLessons”, I couldn't refuse.

The tutorial starts from the basics showing how to install Python and its data analysis libraries. Then it continues explaining the main uses that data scientists and engineers practice during their analysis: importing and cleaning data, vectorial computing, visualization and data summarization.

Most of the videos are commented sessions of IPython notebook sometimes supported by some slides. The authors go deep into the explanation of how to use the libraries for the manipulation of the data (Numpy, Scipy and Pandas), while they summarize the potential of the other complementary libraries. In particular, the last video is a survey of various visualization tools.

In conclusion, this video tutorial provides a solid introduction to the main tools for data analysis in Python and a clear view of the open source Python tools relevant to scientific and engineering programming. This tutorial seems perfect for people who need to learn the technical methodologies for data analysis and for people who already know Python but want to acquire skills about data analysis.

Thursday, December 12, 2013

Multiple axes and subplots in Plotly

Some time ago we have seen how to visualize 2D histograms with Plotly and in this post we will see how to use one of the mostin interesting new features introduced by the Plotly guys: multiple axes into subplots. This features makes us able to couple subplots, so when you zoom or pan in one subplot, it zooms and pans in the other subplots. Just like the graphs produced by D3.
Here's how to plot a subplots matrix where each cell is a scatterplot between the features of the Iris dataset that we already used here. The first thing we need to do is to convert the data in the format required by the Plotly API:
from sklearn.datasets import load_iris
iris = load_iris()

attr = [f.replace(' (cm)', '') for f in iris.feature_names]
colors = {'setosa': 'rgb(31, 119, 180)', 
          'versicolor': 'rgb(255, 127, 14)', 
          'virginica': 'rgb(44, 160, 44)'}

data = []
for i in range(4):
    for j in range(4):
        for t,flower in enumerate(iris.target_names):
            data.append({"name": flower, 
                         "x":[ == t,i],
                         "y":[ == t,j],
                         "type":"scatter", "mode":"markers",
                         'marker': {'color': colors[flower], 
                         "xaxis": "x"+(str(i) if i!=0 else ''),
                         "yaxis": "y"+(str(j) if j!=0 else '')})
Then, we create a layout to adjust the look and feel:
d = 0.04; # padding
dms = [[i*d+i*(1-3*d)/4,i*d+((i+1)*(1-3*d)/4)] for i in range(4)]

layout = {
    "xaxis":{"domain":dms[0], "title":attr[0], 
    "yaxis":{"domain":dms[0], "title":attr[0], 
    "xaxis1":{"domain":dms[1], "title":attr[1], 
    "yaxis1":{"domain":dms[1], "title":attr[1], 
    "xaxis2":{"domain":dms[2], "title":attr[2], 
    "yaxis2":{"domain":dms[2], "title":attr[2], 
    "xaxis3":{"domain":dms[3], "title":attr[3], 
    "yaxis3":{"domain":dms[3], "title":attr[3], 
    "width": 500,
    "height": 550,
    "title":"Iris flower data set",
    "titlefont":{'color':'rgb(67,67,67)', 'size': 20}
Finally, we import the plotly module (see this page for more details about the installation) and we are read to invoke the Plotly remote service:
import plotly
p = plotly.plotly('supersexyusername', 'mysecretkey')
# iplot shows the graph in the ipython notebook
# use plot if you're outside of the notebook
p.iplot(data,layout=layout, width=500,height=550)
The result should be as follows: This interactive graph of the iris data set below was inspired by this wonderful D3 example by Mike Bostock. Find out more example of Plotly visualizations in Python inside the IPython notebook here.