DataFu

DataFu is a collection of user-defined functions for working with large-scale data in Hadoop and Pig. This library was born out of the need for a stable, well-tested library of UDFs for data mining and statistics. It is used at LinkedIn in many of our off-line workflows for data derived products like "People You May Know" and "Skills". It contains functions for:

• PageRank
• Quantiles (median), variance, etc.
• Sessionization
• Convenience bag functions (e.g., set operations, enumerating bags, etc)
• Convenience utility functions (e.g., assertions, easier writing of EvalFuncs)
• and more...

Each function is unit tested and code coverage is being tracked for the entire library. It has been tested against pig 0.10.

http://data.linkedin.com/opensource/datafu

What can you do with it?

Here's a taste of what you can do in Pig.

Statistics

Compute the median of sequence of sorted bags:

define Median datafu.pig.stats.Median();

-- input: 3,5,4,1,2
input = LOAD 'input' AS (val:int);

grouped = GROUP input ALL;

-- produces median of 3
medians = FOREACH grouped {
  sorted = ORDER input BY val;
  GENERATE Median(sorted.val);
}

Similarly, compute any arbitrary quantiles:

define Quantile datafu.pig.stats.Quantile('0.0','0.5','1.0');

-- input: 9,10,2,3,5,8,1,4,6,7
input = LOAD 'input' AS (val:int);

grouped = GROUP input ALL;

-- produces: (1,5.5,10)
quantiles = FOREACH grouped {
  sorted = ORDER input BY val;
  GENERATE Quantile(sorted.val);
}

Set Operations

Treat sorted bags as sets and compute their intersection:

define SetIntersect datafu.pig.bags.sets.SetIntersect();

-- ({(3),(4),(1),(2),(7),(5),(6)},{(0),(5),(10),(1),(4)})
input = LOAD 'input' AS (B1:bag{T:tuple(val:int)},B2:bag{T:tuple(val:int)});

-- ({(1),(4),(5)})
intersected = FOREACH input {
  sorted_b1 = ORDER B1 by val;
  sorted_b2 = ORDER B2 by val;
  GENERATE SetIntersect(sorted_b1,sorted_b2);
}

Compute the set union:

define SetUnion datafu.pig.bags.sets.SetUnion();

-- ({(3),(4),(1),(2),(7),(5),(6)},{(0),(5),(10),(1),(4)})
input = LOAD 'input' AS (B1:bag{T:tuple(val:int)},B2:bag{T:tuple(val:int)});

-- ({(3),(4),(1),(2),(7),(5),(6),(0),(10)})
unioned = FOREACH input GENERATE SetUnion(B1,B2);

Operate on several bags even:

intersected = FOREACH input GENERATE SetUnion(B1,B2,B3);

Bag operations

Concatenate two or more bags:

define BagConcat datafu.pig.bags.BagConcat();

-- ({(1),(2),(3)},{(4),(5)},{(6),(7)})
input = LOAD 'input' AS (B1: bag{T: tuple(v:INT)}, B2: bag{T: tuple(v:INT)}, B3: bag{T: tuple(v:INT)});

-- ({(1),(2),(3),(4),(5),(6),(7)})
output = FOREACH input GENERATE BagConcat(B1,B2,B3);

Append a tuple to a bag:

define AppendToBag datafu.pig.bags.AppendToBag();

-- ({(1),(2),(3)},(4))
input = LOAD 'input' AS (B: bag{T: tuple(v:INT)}, T: tuple(v:INT));

-- ({(1),(2),(3),(4)})
output = FOREACH input GENERATE AppendToBag(B,T);

PageRank

Run PageRank on a large number of independent graphs:

define PageRank datafu.pig.linkanalysis.PageRank('dangling_nodes','true');

topic_edges = LOAD 'input_edges' as (topic:INT,source:INT,dest:INT,weight:DOUBLE);

topic_edges_grouped = GROUP topic_edges by (topic, source) ;
topic_edges_grouped = FOREACH topic_edges_grouped GENERATE
  group.topic as topic,
  group.source as source,
  topic_edges.(dest,weight) as edges;

topic_edges_grouped_by_topic = GROUP topic_edges_grouped BY topic; 

topic_ranks = FOREACH topic_edges_grouped_by_topic GENERATE
  group as topic,
  FLATTEN(PageRank(topic_edges_grouped.(source,edges))) as (source,rank);

skill_ranks = FOREACH skill_ranks GENERATE
  topic, source, rank;

This implementation stores the nodes and edges (mostly) in memory. It is therefore best suited when one needs to compute PageRank on many reasonably sized graphs in parallel.

Start Using It

The JAR can be found here in the Maven central repository. The GroupId and ArtifactId are com.linkedin.datafu and datafu, respectively.

If you are using Ivy:

<dependency org="com.linkedin.datafu" name="datafu" rev="0.0.4"/>

If you are using Maven:

<dependency>
  <groupId>com.linkedin.datafu</groupId>
  <artifactId>datafu</artifactId>
  <version>0.0.4</version>
</dependency>

Or you can download one of the packages from the downloads section.

Working with the source code

Here are some common tasks when working with the source code.

Build the JAR

ant jar

Run all tests

ant test

Run specific tests

Override testclasses.pattern, which defaults to **/*.class. For example, to run all tests defined in QuantileTests:

ant test -Dtestclasses.pattern=**/QuantileTests.class

Compute code coverage

ant coverage

Contribute

The source code is available under the Apache 2.0 license.

For help please see the discussion group. Bugs and feature requests can be filed here.