def message(self, msg):
"""Message receive function for show_pairs toggling and reset"""
vocab = self.vocab_out
# Send the new labels
if self.config.show_pairs:
vocab.include_pairs = True
if isinstance(vocab, spa.Vocabulary):
self.data.append(
'["reset_legend_and_data", "%s"]' % (
'","'.join(vocab.keys + vocab.key_pairs)))
# if we're starting to show pairs, track pair length
self.old_pairs_length = len(vocab.key_pairs)
else:
self.data.append(
'["reset_legend_and_data", "%s"]' % (
'","'.join(set(vocab.keys()) | pairs(vocab))))
# if we're starting to show pairs, track pair length
self.old_pairs_length = len(pairs(vocab))
else:
vocab.include_pairs = False
if isinstance(vocab, spa.Vocabulary):
self.data.append('["reset_legend_and_data", "%s"]'
% ('","'.join(vocab.keys)))
else:
self.data.append('["reset_legend_and_data", "%s"]'
% ('","'.join(vocab)))
def gather_data(self, t, x):
vocab = self.vocab_out
if isinstance(vocab, spa.Vocabulary):
length = len(vocab.keys)
else:
length = len(vocab)
if self.old_vocab_length != length:
self.update_legend(vocab)
# get the similarity and send it
key_similarity = np.dot(vocab.vectors, x)
simi_list = ['{:.2f}'.format(simi) for simi in key_similarity]
if self.config.show_pairs:
# briefly there can be no pairs, so catch the error
try:
if isinstance(vocab, spa.Vocabulary):
pair_similarity = np.dot(vocab.vector_pairs, x)
else:
pair_similarity = (np.dot(vocab.parse(p).v, x) for p in pairs(vocab))
simi_list += ['{:.2f}'.format(simi) for simi in pair_similarity]
except TypeError:
pass
if(simi_list != []):
self.data.append( '["data_msg", %g, %s]'
%( t, ",".join(simi_list) ) )
def gather_data(self, t, x):
vocab = self.vocab_out
key_similarities = np.dot(vocab.vectors, x)
over_threshold = key_similarities > 0.01
if isinstance(vocab, spa.Vocabulary):
matches = zip(key_similarities[over_threshold],
np.array(vocab.keys)[over_threshold])
if self.config.show_pairs:
self.vocab_out.include_pairs = True
pair_similarities = np.dot(vocab.vector_pairs, x)
over_threshold = pair_similarities > 0.01
pair_matches = zip(pair_similarities[over_threshold],
np.array(vocab.key_pairs)[over_threshold])
matches = itertools.chain(matches, pair_matches)
else:
matches = zip(
key_similarities[over_threshold],
[k for i, k in enumerate(vocab) if over_threshold[i]])
if self.config.show_pairs:
pair_similarities = np.array(
[np.dot(vocab.parse(p).v, x) for p in pairs(vocab)])
over_threshold = pair_similarities > 0.01
pair_matches = zip(
pair_similarities[over_threshold],
(k
for i, k in enumerate(pairs(vocab)) if over_threshold[i]))
matches = itertools.chain(matches, pair_matches)
text = ';'.join(
['%0.2f%s' % (min(sim, 9.99), key) for sim, key in matches])
# msg sent as a string due to variable size of pointer names
msg = '%g %s' % (t, text)
self.data.append(msg)
if self.override_target is None:
return np.zeros(self.vocab_out.dimensions)
else:
v = (self.override_target.v - x) * 3
return v
def update_legend(self, vocab):
# pass all the missing keys
legend_update = []
if isinstance(vocab, spa.Vocabulary):
legend_update += (vocab.keys[self.old_vocab_length:])
self.old_vocab_length = len(vocab.keys)
else:
legend_update += (list(vocab.keys())[self.old_vocab_length:])
self.old_vocab_length = len(vocab)
# and all the missing pairs if we're showing pairs
if self.config.show_pairs:
# briefly there can be no pairs, so catch the error
try:
key_pairs = list(pairs(vocab))
legend_update += key_pairs[self.old_pairs_length:]
self.old_pairs_length = len(key_pairs)
except TypeError:
pass
self.data.append('["update_legend", "%s"]'
%('","'.join(legend_update)))
def test_pairs():
v = Vocabulary(64)
v.populate('A; B; C')
actual = pairs(v)
expected = {'A*B', 'A*C', 'B*C'}
assert actual == expected
def test_pairs():
v = Vocabulary(64)
v.populate("A; B; C")
actual = pairs(v)
expected = {"A*B", "A*C", "B*C"}
assert actual == expected