Get fresh updates from Hortonworks by email

Once a month, receive latest insights, trends, analytics information and knowledge of Big Data.


Get Started


Ready to Get Started?

Download sandbox

How can we help you?

closeClose button
October 01, 2012
prev slideNext slide

Pig Macro for TF-IDF Makes Topic Summarization 2 Lines of Pig

In a recent post we used Pig to summarize documents via the Term-Frequency, Inverse Document Frequency (TF-IDF) algorithm.

In this post, we’re going to turn that code into a Pig macro that can be called in one line of code:

import 'tfidf.macro';
my_tf_idf_scores = tf_idf(id_body, 'message_id', 'body');

Our macro, in filename tfidf.macro looks just like our pig script, with a couple of new lines. Note the use of macro variables for input and output preceded with the ‘$’ character: $in_relation, $out_relation, $id_field and $text_field. These let us apply the variable to any relation with a unique identifier field and a text body field. You can get it on github here. The file which tests the macro is here. The code that the macro generates is here.

DEFINE tf_idf(in_relation, id_field, text_field) RETURNS out_relation {
  token_records = foreach $in_relation generate $id_field, FLATTEN(TOKENIZE($text_field)) as tokens;
  /* Calculate the term count per document */
  doc_word_totals = foreach (group token_records by ($id_field, tokens)) generate 
    FLATTEN(group) as ($id_field, token), 
    COUNT_STAR(token_records) as doc_total;
  /* Calculate the document size */
  pre_term_counts = foreach (group doc_word_totals by $id_field) generate
    group AS $id_field,
    FLATTEN(doc_word_totals.(token, doc_total)) as (token, doc_total), 
    SUM(doc_word_totals.doc_total) as doc_size;
  /* Calculate the TF */
  term_freqs = foreach pre_term_counts generate $id_field as $id_field,
    token as token,
    ((double)doc_total / (double)doc_size) AS term_freq;
  /* Get count of documents using each token, for idf */
  token_usages = foreach (group term_freqs by token) generate
    FLATTEN(term_freqs) as ($id_field, token, term_freq),
    COUNT_STAR(term_freqs) as num_docs_with_token;
  /* Get document count */
  just_ids = foreach $in_relation generate $id_field;
  ndocs = foreach (group just_ids all) generate COUNT_STAR(just_ids) as total_docs;
  /* Note the use of Pig Scalars to calculate idf */
  $out_relation = foreach token_usages {
    idf    = LOG((double)ndocs.total_docs/(double)num_docs_with_token);
    tf_idf = (double)term_freq * idf;
    generate $id_field as $id_field,
      token as score,
      (chararray)tf_idf as value:chararray;

Note that to debug macros, we can use the -r flag, which will expand the code the macro generates into a .expanded file. It is worth pointing out that this takes us from 37 lines of Pig to 2 lines of pig. Macros facilitate code modularization, re-use and sharing.

Are you sharing enough Hadoop code in your enterprise?


Leave a Reply

Your email address will not be published. Required fields are marked *

If you have specific technical questions, please post them in the Forums

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>