Scoring scripts for Data Science Week 2015 Datathon in Moscow

badass-checker.py

#!/usr/bin/env python

import sys
import json

team_path = '/users/team%s/ozon_recoms.txt' % sys.argv[1]

badass_detected = False


with open(team_path) as f:
    for line in f:
        data = json.loads(line)

        tmrs = data['recoms']

        if len(tmrs.keys()) > 1000:
            badass_detected = True
            break


if badass_detected:
    print('Badass detected!')
else:
    print('OK')

ndcg_scoring.py

#!/usr/bin/env python

import sys
import json
import math

true_path = 'true_recoms.txt'
team_path = 'recoms.txt'

result = { 'score': 0, 'details': 'failed' }

true_recs = {}

with open(true_path) as f:
    for line in f:
        data = json.loads(line)
        true_recs[data['item']] = data['true_recoms']

ndcg_sum = 0
lines_number = 0

with open(team_path) as f:
    for line in f:
        data = json.loads(line)
        lines_number += 1

        trs = true_recs[data['item']]
        tmrs = data['recoms']

        sorted_trs = sorted(trs.items(), key = lambda x: float(x[1]), reverse = True)
        sorted_tmrs = sorted(tmrs.items(), key = lambda x: float(x[1]), reverse = True)

        dcg = 0
        idcg = 0

        for i in range(len(trs)):
            delta = sorted_trs[i][1]
            if i + 1 != 1:
                delta = delta / float(math.log(i + 1, 2))
            idcg += delta

        for i in range(len(tmrs)):
            if sorted_tmrs[i][0] in trs:
                delta = trs[sorted_tmrs[i][0]]
                if i + 1 != 1:
                    delta = delta / float(math.log(i + 1, 2))
                dcg += delta

        ndcg_sum += dcg / idcg

    if lines_number != len(true_recs.keys()):
        result['details'] = 'Wrong lines number'
    else:
        result['details'] = 'ok'
        result['score'] = ndcg_sum / lines_number


print repr(result)

rmse_scoring.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import sys
import json
import operator
import math

from itertools import izip


true_path = 'true_hh_salaries.txt'

team_path = 'hh_salaries.txt'

result = { 'score': 1, 'details': 'failed' }

to_error = 0
from_error = 0

to_lines_number = 0
to_total = 0
from_lines_number = 0
from_total = 0

for true_line, team_line in izip(open(true_path), open(team_path)):
    trdata = json.loads(true_line)
    tmdata = json.loads(team_line)

    if trdata['id'] != tmdata['id']:
        raise('Ids mismatch error!')
        break

    if trdata['salary']['to']:
        to_total += 1
        if tmdata['salary']['to']:
            to_lines_number += 1
            trto = float(trdata['salary']['to'])
            tmto = float(tmdata['salary']['to'])
            to_error += abs(min(trto, tmto) / max(trto, tmto) - 1)**2

    if trdata['salary']['from']:
        from_total += 1
        if tmdata['salary']['from']:
            from_lines_number += 1
            trfrom = float(trdata['salary']['from'])
            tmfrom = float(tmdata['salary']['from'])
            from_error += abs(min(trfrom, tmfrom) / max(trfrom, tmfrom) - 1)**2

predicted_to = to_lines_number / float(to_total)
predicted_from = from_lines_number / float(from_total)

if predicted_to >= 0.5 or predicted_from >= 0.5:
    to_error = math.sqrt(to_error / to_lines_number)
    from_error = math.sqrt(from_error / from_lines_number)

    result['score'] = (to_error + from_error) / 2
    result['details'] = 'Average relative error for "to" – %s, "from" – %s. Predicted "to" lines – %s, "from" – %s' % (to_error, from_error, predicted_to, predicted_from)
else:
    result['details'] = 'Not enough to/from lines predicted'

print repr(result)