# WARNING: spacy's default English model *does not provide* word vector data.
# Instead, it provides "good guess" vectors that are lower quality.
# Download a larger language model, like this, to receive vector data:
#   python -m spacy download en_core_web_lg

# Inputs: document_string (a str)

import numpy as np
import spacy

# Load a language model and parse your document.
nlp = spacy.load('en_core_web_lg')
doc = nlp(document_string)

# Retrieve word vectors for each token.
for token in doc:
    print('Vector for %s:' % token, token.vector)

# Compute word similarities using vector data.
# We'll put these into a numpy matrix.
matrix_rows = []

for token1 in doc:
    row = [token1.similarity(token2) for token2 in doc]
    matrix_rows.append(row)

similarity_matrix = np.array(matrix_rows)

import spacy


# Set up functions to help produce human-friendly printing.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

def skip_and_print(*args):
    """ Act like print(), but skip a line before printing. """
    print('\n' + str(args[0]), *args[1:])

# The next definitions work together to help print tables.

col_widths = []

def row(*values):
    """ Return a str with `values` spaced by `col_widths`. """
    fmt = ' '.join('%%%ds' % width for width in col_widths)
    return fmt % tuple(values)

def hr():
    """ Return a horizontal rule str using `col_widths`. """
    size = sum(map(abs, col_widths)) + len(col_widths) - 1
    return '~' * size


# Load a language model and parse a document.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

# spacy's default English language model doesn't have word
# vector data, so we must use a larger variant to use vectors.
# Here's how to download a large model for English:
#  python -m spacy download en_core_web_lg
nlp = spacy.load('en_core_web_lg')
document_string = 'happy glad cheddar munster'

skip_and_print('Working with string: "%s"' % document_string)
doc = nlp(document_string)


# Retrieve word vectors for each token.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

skip_and_print('Word vectors for each token:\n')

col_widths = [7, 1, -30]
print(row('Word', '|', 'Vector'))
print(hr())

for token in doc:
    v = token.vector
    v_str = ('%5.2f ' * 4) % tuple(v[:4])
    print(row(token, '|', '(' + v_str + '...)'))

# Word vectors for each token:
#
#    Word | Vector
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#   happy | ( 0.04  0.41 -0.52 -0.07 ...)
#    glad | ( 0.07  0.25 -0.53 -0.03 ...)
# cheddar | (-0.63  0.53  0.23 -0.16 ...)
# munster | (-0.15  0.76  0.48  0.23 ...)


# Compute word similarities using vector data.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

skip_and_print('Similarity matrix for the above words:\n')

col_widths = [7] * 5  # Our longest word has length 7.
print(row('', *doc))
print(hr())

for token1 in doc:
    values = [token1]
    for token2 in doc:
        values.append('%5.2f' % token1.similarity(token2))
    print(row(*values))

# Similarity matrix for the above words:
# 
#           happy    glad cheddar munster
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#   happy    1.00    0.77    0.06   -0.01
#    glad    0.77    1.00    0.07    0.00
# cheddar    0.06    0.07    1.00    0.21
# munster   -0.01    0.00    0.21    1.00