Genre classification test

compare.py

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

import csv
import os
from collections import defaultdict, Counter

comparisons = defaultdict(dict)
true_genres = {}
test_genres = {}
genre_confusion = defaultdict(Counter)

def nn(trackname):
    neighbors = comparisons[trackname].keys()
    return min(neighbors, key=lambda x: comparisons[trackname][x])


def parse_csv(filename, parse_func):
    with file(filename) as f:
        dialect = csv.Sniffer().sniff(f.read(1024))
        f.seek(0)
        reader = csv.reader(f, dialect)
        for row in reader:
            parse_func(row)

# parse CSVs

def parse_comparison_row(row):
    comparisons[row[0]][row[1]] = float(row[4])
    comparisons[row[1]][row[0]] = float(row[4])


def parse_tracklist_row(row):
    true_genres[os.path.basename(row[5])] = row[0]

def parse_groundtruth_row(row):
    true_genres[os.path.basename(row[0])] = row[1]

parse_csv('comparisons.csv', parse_comparison_row)
parse_csv('tracklist.csv', parse_tracklist_row)
parse_csv('ground_truth.csv', parse_groundtruth_row)

for track in true_genres.keys():
    try:
        track_nn = nn(track)
    except SystemExit, KeyboardInterrupt:
        sys.exit(1)
    except:
        print track
    true_genre = true_genres[track]
    test_genre = true_genres[track_nn]

    test_genres[track] = test_genre
    genre_confusion[true_genre][test_genre] += 1

# assess accuracy

correct = 0
all_items = 0

for (genre, genre_row) in genre_confusion.iteritems():
    correct += genre_confusion[genre][genre]
    all_items += sum(genre_row.values())

print '{:02%} correct'.format(float(correct) / all_items)