Have you ever wanted to know what the most mean color is? Red might come to mind — a fiery and belligerent color. But I’ll do one better! I’ll give you the color vector that’s closest to every other color vector.

Here, the most average and mean color can be determined with word2vec and vector averaging.

Process

1. Load the word2vec model
2. Load the list of colors
3. Sum the vectors of each color together
4. Divide by the number of colors
5. Find the color vector with the largest cosine similarity to that vector

Python

pip install gensim

import numpy
import gensim.downloader

queen_values = {}

for name in gensim.downloader.info()['models'].keys():
    try:
        model = gensim.downloader.load(name)
        t = model.most_similar(positive=['king', 'woman'], negative=['man'])
        for n, v in t:
            if n == 'queen':
                queen_values[name] = v
        print(t[0])
    except:
        pass

('queen', 0.7786749005317688)
('queen', 0.7118193507194519)
('queen', 0.8523604273796082)
('queen', 0.7698540687561035)
('queen', 0.6978678107261658)
('queen', 0.6713277101516724)
('meets', 0.8841924071311951)
('prince', 0.759779691696167)
('queen', 0.7052316069602966)
('queen', 0.6820898056030273)

best_model = sorted(queen_values, key=queen_values.get, reverse=True)[0]
print('\n', best_model)
model = gensim.downloader.load(best_model)

 glove-wiki-gigaword-50

colors = [
    'white', 'black', 'silver',
    'red', 'maroon', 'purple',
    'orange', 'brown', 'beige',
    'yellow', 'gold', 'tan',
    'green', 'lime', 'olive',
    'blue', 'cyan', 'teal',
    'indigo', 'lavender', 'pink',
    'violet', 'magenta', 'fuchsia',
]

# start by averaging grays
avg_vector = (model['gray'] + model['grey']) / 2

for color in colors:
    avg_vector += model[color]

# divide by the number of colors
avg_vector /= (len(colors) + 1)

avg_colors = {}

for color in colors:
    sim = numpy.dot(avg_vector, model[color]) / \
        (numpy.linalg.norm(avg_vector) * numpy.linalg.norm(model[color]))
    avg_colors[color] = sim

for color in sorted(avg_colors, key=avg_colors.get, reverse=True):
    print(color, '\t\t', avg_colors[color])

purple		0.92791396
pink		0.9113098
blue		0.9093884
yellow		0.88690674
red		0.872676
green		0.8654766
orange		0.84672266
black		0.83403736
white		0.8287787
lavender	0.81729794
maroon		0.7757962
fuchsia		0.7409902
brown		0.7358376
olive		0.73582804
magenta		0.68170786
violet		0.6786235
silver		0.6677471
beige		0.66110957
lime		0.6299113
indigo		0.5789356
gold		0.56747013
tan		0.5436225
teal		0.517831
cyan		0.5167587

Conclusion

There isn’t any purpose for this post, but it’s interesting to know that the meanest color is purple. I mostly wanted to start thinking about fun ways to use word2vec. I think I’d like to make a tool to assist writers in some way. I’m not sure what I’ll do with it, but I’m sure it’ll be fun.