Rosetta Code

A collection of MapReduce tasks translated (from Pig, Hive, MapReduce streaming, etc.) into Scalding. For fully runnable code, see the repository here.

Word Count

Hadoop Streaming (Ruby)

# Emit (word, count) pairs.
def mapper
  STDIN.each_line do |line|
    line.split.each do |word|
      puts [word, 1].join("\t")
    end
  end
end

# Aggregate all (word, count) pairs for a particular word.
#
# In Hadoop Streaming (unlike standard Hadoop), the reducer receives
# rows from the mapper *one at a time*, though the rows are guaranteed
# to be sorted by key (and every row associated to a particular key
# will be sent to the same reducer).
def reducer
  curr_word = nil
  curr_count = 0
  STDIN.each_line do |line|
    word, count = line.strip.split("\t")
    if word != curr_word
      puts [curr_word, curr_count].join("\t")
      curr_word = word
      curr_count = 0
    end
    curr_count += count.to_i
  end
  
  puts [curr_word, curr_count].join("\t") unless curr_word.nil?
end

Cascalog

; Takes a single piece of text as input.
; Outputs a tuple for each word in the text.
(defmapcatop tokenize [text]
  (seq (.split text "\\s+")))

(defn word-count [input-filename]
	(let [input (hfs-textline input-filename)]
    (<- [?word ?count]
        (input ?textline)
        (tokenize ?textline :> ?word)
        (c/count ?count))))

(?- (stdout) (word-count "tweets.tsv"))

Hive

# tokenizer.py
import sys

for line in sys.stdin:
  for word in line.split():
    print word

CREATE TABLE tweets (text STRING);
LOAD DATA LOCAL INPATH 'tweets.tsv' OVERWRITE INTO TABLE tweets;

SELECT word, COUNT(*) AS count
FROM (
  SELECT TRANSFORM(text) USING 'python tokenizer.py' AS word
  FROM tweets
) t
GROUP BY word;

Pig

tweets = LOAD 'tweets.tsv' AS (text:chararray);
words = FOREACH tweets GENERATE FLATTEN(TOKENIZE(text)) AS word;
word_groups = GROUP words BY word;
word_counts = FOREACH word_groups GENERATE group AS word, COUNT(words) AS count;

STORE word_counts INTO 'word_counts.tsv';

Scalding

Tsv("tweets.tsv", 'text)
  .flatMap('text -> 'word) { text : String => text.split("\\s+") }
  .groupBy('word) { _.size }
  .write(Tsv("word_counts.tsv"))

Distributed Grep

Pig

%declare PATTERN '.*hello.*';

tweets = LOAD 'tweets.tsv' AS (text:chararray);
results = FILTER tweets BY (text MATCHES '$PATTERN');

Scalding

val Pattern = ".*hello.*";

Tsv("tweets.tsv", 'text)
  .filter('text) { text : String => text.matches(Pattern) }

Inverted Index

Pig

tweets = LOAD 'tweets.tsv' AS (tweet_id:int, text:chararray);

words = FOREACH tweets GENERATE tweet_id, FLATTEN(TOKENIZE(text)) AS word;
word_groups = GROUP words BY word;
inverted_index = FOREACH word_groups GENERATE group AS word, words.tweet_id;

Scalding

val tweets = Tsv("tweets.tsv", ('id, 'text))

val wordToTweets =
  tweets
    .flatMap(('id, 'text) -> ('word, 'tweetId)) { 
      fields : (Long, String) => 
      val (tweetId, text) = fields
      text.split("\\s+").map { word => (word, tweetId) }
    }

val invertedIndex =
  wordToTweets.groupBy('word) {  _.toList[Long]('tweetId -> 'tweetIds) }