def make_colorizer():
from csc.divisi2 import examples
log.info('Loading ConceptNet matrix')
sa = examples.spreading_activation()
log.info('Loading color matrix')
colors = make_color_matrix()
log.info('Building colorizer')
return Colorizer(sa, colors)
def training_and_test_data():
print "Constructing test/train from xkcd"
train = defaultdict(list)
test = defaultdict(list)
combined = defaultdict(list)
with open('grouped_color_data.txt') as inputlines:
for line in inputlines:
try:
colorname, userid, r, g, b, monitor, colorblind, male = line.strip().split('|')
except ValueError:
continue
rgblist = [float(r), float(g), float(b)]
combined[colorname].append(rgblist)
print colorname.decode('utf-8').encode('ascii', 'replace')
#for colorname in combined.keys():
# if len(combined[colorname]) < 3:
# del combined[colorname]
from csc.divisi2.examples import spreading_activation
spread = spreading_activation()
import random
possible_concepts = set(spread.row_labels) & set(combined)
test_concepts = random.sample(possible_concepts, 200)
out = open('test_concepts.txt', 'w')
for concept in test_concepts:
print "Test concept:", concept
print >> out, concept
out.close()
training = True
for colorname in combined:
concepts = en.nl.extract_concepts(colorname, check_conceptnet=True)
censored = False
for concept in concepts:
if concept in test_concepts:
censored = True
if censored: target = test
else: target = train
target[colorname] = combined[colorname]
return train, test
def run_leave_n_out():
from csc.divisi2 import examples
import colors
log.info('Loading ConceptNet matrix')
sa = examples.spreading_activation()
log.info('Loading test input')
train_input, test_input = training_and_test_data()
log.info('Loading test data')
test = make_test_data()
log.info('Building colorizer')
cmatrix = make_color_matrix()
colorizer = Colorizer(sa, cmatrix)
dist_dict = {}
baseline_dict = {}
distances = {
'baseline': [],
'weighted': [],
'inter_annotator': [],
}
test_concepts = set(test.keys()) & set(sa.row_labels)
for colorname in test_concepts:
try:
labout = tuple(colorizer.lab_color_for_concept(colorname)[:3])
except TypeError:
continue
labact = test[colorname]
rgbact = lab_to_rgb(labact)
dist = euclid(labout,labact)
distances['weighted'].append(dist)
baseline = euclid([50, 0, 0], labact)
distances['baseline'].append(baseline)
#try:
# wnout = tuple(colorizer.lab_color_for_concept_wordnet(colorname)[:3])
# wndist = euclid(wnout, labact)
# distances['wordnet'].append(wndist)
# rgbwnout = lab_to_rgb(wnout)
# print colorname, '(wordnet)', rgbact, rgbwnout, str(wndist)
#except TypeError:
# pass
#try:
# prismdata = [rgb_to_lab([color.__r, color.__g, color.__b]) for color in colors.prism(colorname)]
# prism_avg = numpy.mean(numpy.array(prismdata), axis=0)
# prism_dist = euclid(prism_avg, labact)
# distances['nodebox_prism'].append(prism_dist)
#except ZeroDivisionError:
# pass
inter_annotator = euclid(rgb_to_lab(test_input[colorname][0]),
labact)
if inter_annotator == 0:
inter_annotator = euclid(rgb_to_lab(test_input[colorname][1]),
labact)
distances['inter_annotator'].append(inter_annotator)
rgbout = lab_to_rgb(labout)
print colorname, rgbact, rgbout, str(dist)
totals = {}
for key, values in distances.items():
totals[key] = sum(values)/len(values)
totals['total'] = len(distances['weighted'])
print totals
return totals
