statistics.py

# coding=utf-8

import os, sys
import numpy as np
import csv
import re

artist_labels = {}
style_labels = {}
genre_labels = {}
date_labels = {} #global so can be used elsewhere
max_examples = 1200 #maximum number of examples for any given label
min_examples = 600 #TODO: make this variable

# all have +600 examples
date_check = [
	'1910-1919', '1900-1909', '1890-1899', '1880-1889', '1920-1929', 
	'1930-1939', '1870-1879', '1960-1969',  '1950-1959', '1940-1949', 
	'1970-1979', '1860-1869', '1980-1989',  '1850-1859', '1990-1999',
	'1840-1849',  '1830-1839',  '1820-1829',  '1810-1819' 
]

genre_check = [						#number of examples
	'portrait', 		 			#12926
	'landscape', 		 			#11548
	'genre painting', 		 		#10984
	'abstract', 		 			#7201
	'religious painting', 		 	#5703
	'cityscape', 		 			#4089
	'sketch and study', 		 	#2778
	'illustration', 		 		#2493
	'still life', 		 			#2464
	'symbolic painting', 		 	#1959
	'figurative', 		 			#1782
	'nude painting (nu)', 		 	#1758
	'design', 		 				#1577
	'mythological painting', 		 #1493
	'marina', 		 				#1385
	'flower painting', 		 		#1270
	'animal painting', 		 		#1233
	'self-portrait', 		 		#1199
	'allegorical painting', 		 #809
	'history painting'				#656
]


style_check = [						#number of examples
	'Impressionism', 		 		#8220
	'Realism', 		 				#8112
	'Romanticism', 		 			#7041
	'Expressionism', 		 		#5325
	'Post-Impressionism' 		 	#4527

	
	#'Art Nouveau (Modern)', 		#3779
	#'Baroque', 					#3254
	#'Surrealism', 					#3133
	#'Symbolism', 		 			#2626
	#'Rococo', 		 				#2101
	#'Northern Renaissance', 		#1824
	#'Naïve Art (Primitivism)',		#1776
	#'Neoclassicism', 		 		#1622
	#'Abstract Expressionism', 		#1546
	#'Cubism', 		 				#1316
	#'Ukiyo-e', 		 			#1137
	#'Early Renaissance', 		 	#1052
	#'High Renaissance', 		 	#1050
	#'Mannerism (Late Renaissance)',#1025
	#'Art Informel', 		 		#987
	#'Academicism', 		 		#766
	#'Abstract Art', 		 		#759
	#Magic Realism', 				#696
	#'Color Field Painting', 		#675
	#'Pop Art', 		 			#619

]


def intersect(a, b):
	return list(set(a) & set(b))


def label_mapping(filename):
	"""
	Creates mapping of filename to label from CSV

	Args:
		filename	: name of dataset CSV file
	"""

	


	with open(filename, 'r') as infile:
		reader = csv.reader(infile)
		next(reader, None) # ignore first line since they're column labels

		#filename, artist, title, style, genre, date
		for line in reader:
			img = line[0]
			artist = line[1]
			style = line[3]
			genre = line[4]
			date = re.findall(r'\d+', line[5]) #parse any unwanted stuff

			#img and artist fields always present, no need to check
			artist_labels[img] = artist


			if style != '' and style in style_check:
				#if sum(x == style for x in style_labels.values()) < max_examples: # avoid imbalance
				style_labels[img] = style


			if genre != '' and genre in genre_check:
				#if sum(x == genre for x in genre_labels.values()) < max_examples:
				genre_labels[img] = genre


			if len(date) > 0:
				bucket_len = 10 #buckets of 10 years
				bucket = (int(date[0]) // bucket_len) * bucket_len 
				period = str(bucket) + '-' + str(bucket + (bucket_len - 1))

				if period in date_check:
					#if sum(x == period for x in date_labels.values()) <= max_examples:
					date_labels[img] =  period #parsed_date


def label_stats(label_mapping):
	"""
    Determines the number of exampes for each label, sorted from hightest to lowest

    Args:
        label_mapping : the dictionary label mapping to inspect
    """
	labels = list(label_mapping.values())

	for count, elem in sorted(((labels.count(e), e) for e in set(labels)), reverse=True):
		print('%s: \t\t %d' % (elem, count))

def write_labels(filename, label_mapping):
	#write class labels for future use
	w = csv.writer(open(filename, "w"))
	for key, val in label_mapping.items():
		w.writerow([key, val])

def main():

	label_mapping('train_info.csv')

	'''
	artist_labels
	style_labels
	genre_labels
	date_labels
	'''
	print('STYLE:\n')
	label_stats(style_labels)
	print('=================================================')
	print('GENRE:\n')
	label_stats(genre_labels)
	print('=================================================')
	print('DATES:\n')
	label_stats(date_labels)
	print('=================================================')
	
	#write_labels('train_artist.csv', artist_labels)
	write_labels('train_style.csv', style_labels)
	write_labels('train_genre.csv', genre_labels)
	write_labels('train_date.csv', date_labels)


if __name__ == '__main__':
	main()