MLlib - Frequent Pattern Mining

Mining frequent items, itemsets, subsequences, or other substructures is usually among the first steps to analyze a large-scale dataset, which has been an active research topic in data mining for years. We refer users to Wikipedia’s association rule learning for more information. MLlib provides a parallel implementation of FP-growth, a popular algorithm to mining frequent itemsets.

FP-growth

The FP-growth algorithm is described in the paper Han et al., Mining frequent patterns without candidate generation, where “FP” stands for frequent pattern. Given a dataset of transactions, the first step of FP-growth is to calculate item frequencies and identify frequent items. Different from Apriori-like algorithms designed for the same purpose, the second step of FP-growth uses a suffix tree (FP-tree) structure to encode transactions without generating candidate sets explicitly, which are usually expensive to generate. After the second step, the frequent itemsets can be extracted from the FP-tree. In MLlib, we implemented a parallel version of FP-growth called PFP, as described in Li et al., PFP: Parallel FP-growth for query recommendation. PFP distributes the work of growing FP-trees based on the suffices of transactions, and hence more scalable than a single-machine implementation. We refer users to the papers for more details.

MLlib’s FP-growth implementation takes the following (hyper-)parameters:

minSupport: the minimum support for an itemset to be identified as frequent. For example, if an item appears 3 out of 5 transactions, it has a support of 3/5=0.6.
numPartitions: the number of partitions used to distribute the work.

Examples

FPGrowth implements the FP-growth algorithm. It take a RDD of transactions, where each transaction is an Array of items of a generic type. Calling FPGrowth.run with transactions returns an FPGrowthModel that stores the frequent itemsets with their frequencies. The following example illustrates how to mine frequent itemsets and association rules (see Association Rules for details) from transactions.

import org.apache.spark.rdd.RDD
import org.apache.spark.mllib.fpm.FPGrowth

val data = sc.textFile("data/mllib/sample_fpgrowth.txt")

val transactions: RDD[Array[String]] = data.map(s => s.trim.split(' '))

val fpg = new FPGrowth()
  .setMinSupport(0.2)
  .setNumPartitions(10)
val model = fpg.run(transactions)

model.freqItemsets.collect().foreach { itemset =>
  println(itemset.items.mkString("[", ",", "]") + ", " + itemset.freq)
}

val minConfidence = 0.8
model.generateAssociationRules(minConfidence).collect().foreach { rule =>
  println(
    rule.antecedent.mkString("[", ",", "]")
      + " => " + rule.consequent .mkString("[", ",", "]")
      + ", " + rule.confidence)
}

FPGrowth implements the FP-growth algorithm. It take an JavaRDD of transactions, where each transaction is an Iterable of items of a generic type. Calling FPGrowth.run with transactions returns an FPGrowthModel that stores the frequent itemsets with their frequencies. The following example illustrates how to mine frequent itemsets and association rules (see Association Rules for details) from transactions.

import java.util.Arrays;
import java.util.List;

import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.mllib.fpm.AssociationRules;
import org.apache.spark.mllib.fpm.FPGrowth;
import org.apache.spark.mllib.fpm.FPGrowthModel;

SparkConf conf = new SparkConf().setAppName("FP-growth Example");
JavaSparkContext sc = new JavaSparkContext(conf);

JavaRDD<String> data = sc.textFile("data/mllib/sample_fpgrowth.txt");

JavaRDD<List<String>> transactions = data.map(
  new Function<String, List<String>>() {
    public List<String> call(String line) {
      String[] parts = line.split(" ");
      return Arrays.asList(parts);
    }
  }
);

FPGrowth fpg = new FPGrowth()
  .setMinSupport(0.2)
  .setNumPartitions(10);
FPGrowthModel<String> model = fpg.run(transactions);

for (FPGrowth.FreqItemset<String> itemset: model.freqItemsets().toJavaRDD().collect()) {
  System.out.println("[" + itemset.javaItems() + "], " + itemset.freq());
}

double minConfidence = 0.8;
for (AssociationRules.Rule<String> rule
    : model.generateAssociationRules(minConfidence).toJavaRDD().collect()) {
  System.out.println(
    rule.javaAntecedent() + " => " + rule.javaConsequent() + ", " + rule.confidence());
}

FPGrowth implements the FP-growth algorithm. It take an RDD of transactions, where each transaction is an List of items of a generic type. Calling FPGrowth.train with transactions returns an FPGrowthModel that stores the frequent itemsets with their frequencies.

from pyspark.mllib.fpm import FPGrowth

data = sc.textFile("data/mllib/sample_fpgrowth.txt")

transactions = data.map(lambda line: line.strip().split(' '))

model = FPGrowth.train(transactions, minSupport=0.2, numPartitions=10)

result = model.freqItemsets().collect()
for fi in result:
    print(fi)

Association Rules

AssociationRules implements a parallel rule generation algorithm for constructing rules that have a single item as the consequent.

import org.apache.spark.rdd.RDD
import org.apache.spark.mllib.fpm.AssociationRules
import org.apache.spark.mllib.fpm.FPGrowth.FreqItemset

val freqItemsets = sc.parallelize(Seq(
  new FreqItemset(Array("a"), 15L),
  new FreqItemset(Array("b"), 35L),
  new FreqItemset(Array("a", "b"), 12L)
));

val ar = new AssociationRules()
  .setMinConfidence(0.8)
val results = ar.run(freqItemsets)

results.collect().foreach { rule =>
  println("[" + rule.antecedent.mkString(",")
    + "=>"
    + rule.consequent.mkString(",") + "]," + rule.confidence)
}

AssociationRules implements a parallel rule generation algorithm for constructing rules that have a single item as the consequent.

import java.util.Arrays;

import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.mllib.fpm.AssociationRules;
import org.apache.spark.mllib.fpm.FPGrowth.FreqItemset;

JavaRDD<FPGrowth.FreqItemset<String>> freqItemsets = sc.parallelize(Arrays.asList(
  new FreqItemset<String>(new String[] {"a"}, 15L),
  new FreqItemset<String>(new String[] {"b"}, 35L),
  new FreqItemset<String>(new String[] {"a", "b"}, 12L)
));

AssociationRules arules = new AssociationRules()
  .setMinConfidence(0.8);
JavaRDD<AssociationRules.Rule<String>> results = arules.run(freqItemsets);

for (AssociationRules.Rule<String> rule: results.collect()) {
  System.out.println(
    rule.javaAntecedent() + " => " + rule.javaConsequent() + ", " + rule.confidence());
}

PrefixSpan

PrefixSpan is a sequential pattern mining algorithm described in Pei et al., Mining Sequential Patterns by Pattern-Growth: The PrefixSpan Approach. We refer the reader to the referenced paper for formalizing the sequential pattern mining problem.

MLlib’s PrefixSpan implementation takes the following parameters:

minSupport: the minimum support required to be considered a frequent sequential pattern.
maxPatternLength: the maximum length of a frequent sequential pattern. Any frequent pattern exceeding this length will not be included in the results.
maxLocalProjDBSize: the maximum number of items allowed in a prefix-projected database before local iterative processing of the projected databse begins. This parameter should be tuned with respect to the size of your executors.

Examples

The following example illustrates PrefixSpan running on the sequences (using same notation as Pei et al):

  <(12)3>
  <1(32)(12)>
  <(12)5>
  <6>

PrefixSpan implements the PrefixSpan algorithm. Calling PrefixSpan.run returns a PrefixSpanModel that stores the frequent sequences with their frequencies.

import org.apache.spark.mllib.fpm.PrefixSpan

val sequences = sc.parallelize(Seq(
    Array(Array(1, 2), Array(3)),
    Array(Array(1), Array(3, 2), Array(1, 2)),
    Array(Array(1, 2), Array(5)),
    Array(Array(6))
  ), 2).cache()
val prefixSpan = new PrefixSpan()
  .setMinSupport(0.5)
  .setMaxPatternLength(5)
val model = prefixSpan.run(sequences)
model.freqSequences.collect().foreach { freqSequence =>
println(
  freqSequence.sequence.map(_.mkString("[", ", ", "]")).mkString("[", ", ", "]") + ", " + freqSequence.freq)
}

PrefixSpan implements the PrefixSpan algorithm. Calling PrefixSpan.run returns a PrefixSpanModel that stores the frequent sequences with their frequencies.

import java.util.Arrays;
import java.util.List;

import org.apache.spark.mllib.fpm.PrefixSpan;
import org.apache.spark.mllib.fpm.PrefixSpanModel;

JavaRDD<List<List<Integer>>> sequences = sc.parallelize(Arrays.asList(
  Arrays.asList(Arrays.asList(1, 2), Arrays.asList(3)),
  Arrays.asList(Arrays.asList(1), Arrays.asList(3, 2), Arrays.asList(1, 2)),
  Arrays.asList(Arrays.asList(1, 2), Arrays.asList(5)),
  Arrays.asList(Arrays.asList(6))
), 2);
PrefixSpan prefixSpan = new PrefixSpan()
  .setMinSupport(0.5)
  .setMaxPatternLength(5);
PrefixSpanModel<Integer> model = prefixSpan.run(sequences);
for (PrefixSpan.FreqSequence<Integer> freqSeq: model.freqSequences().toJavaRDD().collect()) {
  System.out.println(freqSeq.javaSequence() + ", " + freqSeq.freq());
}