(変な)おもしろコード見せ合い会

gen.rb

require 'json'

probs = Hash.new { |h,k| h[k] = {} }

STDIN.each do |l|
  d = JSON.load(l.chomp)
  probs[d['l']][d['r']] = (d['prob'] * 1000).to_i
end

def gen(probs)
  left = '@@@START@@@'
  pbs = probs[left]
  generated = []
  loop_num = 0
  while loop_num < 10000 do
    loop_num += 1
    left_c = pbs.map{|r, pb| [r] * pb}.inject(:+).sample
    pbs_c = probs[left_c]
    next if pbs_c.empty?

    left = left_c
    pbs = pbs_c
    break if left == '@@@END@@@'
    generated << left
  end
  return generated
end

loop do
  begin
    code = gen(probs).map{|g| g.gsub('___ENT___', "\n")}.join(' ')
    # 任意のコードを実行するので壊れてもいい環境で実行するべき
    eval(code)
    puts "-----------------------------"
    puts code
  rescue Exception
  end
end

index.sql

CREATE TEMPORARY FUNCTION
  pairs(c STRING)
  RETURNS ARRAY<STRUCT<l STRING,
  r STRING>>
  LANGUAGE js AS """
  const normalized = c.replace(RegExp('\\n+', 'g'), "___ENT___").replace(RegExp(' +', 'g'), " ");
  const splited = normalized.split(' ');
  let left = '@@@START@@@';
  let ps = [];
  splited.forEach(right => {
    ps.push({'l': left, 'r': right});
    left = right;
  })
  ps.push({'l': left, 'r': '@@@END@@@'});
  return ps;
""";
WITH
  token_pairs AS (
  SELECT
    pairs(c.content) ps
  FROM
    `bigquery-public-data.github_repos.sample_contents` c
  JOIN
    `bigquery-public-data.github_repos.sample_files` f
  ON
    c.id = f.id
  WHERE
    ENDS_WITH(f.path, '.rb')
    AND c.content IS NOT NULL ),
  token_freqs AS (
  SELECT
    l,
    r,
    freq
  FROM (
    SELECT
      token.l l,
      token.r r,
      COUNT(1) freq
    FROM
      token_pairs,
      UNNEST (token_pairs.ps) AS token
    GROUP BY
      l,
      r )
  WHERE
    freq >= 5 ),
  left_freqs AS (
  SELECT
    l,
    COUNT(1) freq
  FROM
    token_freqs
  GROUP BY
    l )
SELECT
  tf.l,
  tf.r,
  tf.freq / lf.freq prob
FROM
  token_freqs tf
JOIN
  left_freqs lf
ON
  tf.l = lf.l
ORDER BY
  tf.l