def reorder_numbered_placeholders(input1, input2, by_group=True):
""" Given a stream, identifies placeholders that are suffixed by '_$DIGITS'. These are considered to be
argument placeholders, and will be renumbered with those prefixes in place.
If we are reordering by group, numbering is done by counts in group"""
eid = 0
toks1 = process_sentence(input1)
toks2 = process_sentence(input2)
oldent2newent = {}
group_hist = {}
# First pass scan for entities. Add in replacements
for tok1 in toks1:
matched = is_numbered_placeholder(tok1)
if matched is not None:
if tok1 not in oldent2newent:
if by_group:
if matched not in group_hist:
group_hist[matched] = 0
next_id = group_hist[matched]
group_hist[matched] += 1
else:
next_id = eid
new_entity = matched.format(next_id)
eid += 1
oldent2newent[tok1] = new_entity
# Replace in lang1 and lang2
new_toks1 = [oldent2newent[t] if t in oldent2newent else t for t in toks1]
new_toks2 = [subsfirst(t, oldent2newent) for t in toks2]
rlookup = {}
for oldent, newent in oldent2newent.items():
rlookup[newent] = oldent
return " ".join(new_toks1), " ".join(new_toks2), rlookup
def normalize_sal_entities(input1, input2, ent_prefix="id__"):
""" Get the entity IDs in order, and issues new in-sequence IDs."""
eid = 0
toks1 = process_sentence(input1)
toks2 = process_sentence(input2)
oldent2newent = {}
# First pass scan for entities
for tok1 in toks1:
if is_entity(tok1):
if tok1 not in oldent2newent:
new_entity = "{}{}".format(ent_prefix, eid)
eid += 1
oldent2newent[tok1] = new_entity
elif is_number(tok1):
if tok1 not in oldent2newent:
new_entity = "number__{}".format(eid)
eid += 1
oldent2newent[tok1] = new_entity
# Replace in lang1 and lang2
new_toks1 = [oldent2newent[t] if t in oldent2newent else t for t in toks1]
new_toks2 = [oldent2newent[t] if t in oldent2newent else t for t in toks2]
rlookup = {}
for oldent, newent in oldent2newent.items():
rlookup[newent] = oldent
return " ".join(new_toks1), " ".join(new_toks2), rlookup
def indexes_from_sentence(lang, sentence):
indices = [
lang.word2index.get(word, UNK_token)
for word in process_sentence(sentence)
]
if match_parens:
return indices
else:
return indices + [EOS_token]
def evaluate_pairs(evaluate_fn, eval_pairs):
num_correct = 0
res_str = ""
def emit(str):
nonlocal res_str
res_str += str
res_str += "\n"
return res_str
i_iter = tqdm(range(len(eval_pairs)))
print("Evaluating pairs")
for i in i_iter:
pair = eval_pairs[i]
lang1_str, lang2_str = pair
emit("# {}".format(i))
lang1_toks = process_sentence(lang1_str)
lang2_toks = process_sentence(lang2_str)
num_lang1_toks = len(lang1_toks)
num_lang2_toks = len(lang2_toks)
guessed_words, _ = evaluate_fn(lang1_str)
guessed_sentence = ' '.join(guessed_words)
num_guess_toks = len(guessed_words)
is_correct = exact_match(lang2_toks, guessed_words)
if is_correct:
num_correct += 1
emit('Lang (#={}):\t{}'.format(num_lang1_toks, pair[0]))
emit('Gold (#={}):\t{}'.format(num_lang2_toks, pair[1]))
if is_correct:
emit('Guess (#={}):\t{}'.format(num_guess_toks, guessed_sentence))
else:
emit('* Guess (#={}):\t{}'.format(num_guess_toks, guessed_sentence))
emit('')
acc = num_correct / len(eval_pairs)
emit("Exact acc = {:.5f}".format(acc))
return acc, res_str
def show_attention(input_sentence, output_words, attentions):
# Set up figure with colorbar
fig = plt.figure()
ax = fig.add_subplot(111)
cax = ax.matshow(attentions.numpy(), cmap='bone')
fig.colorbar(cax)
# Set up axes
ax.set_xticklabels([''] + process_sentence(input_sentence) + ['<EOS>'],
rotation=90)
ax.set_yticklabels([''] + output_words)
# Show label at every tick
ax.xaxis.set_major_locator(ticker.MultipleLocator(1))
ax.yaxis.set_major_locator(ticker.MultipleLocator(1))
plt.show()
def indexes_from_sentence(lang, sentence):
return [
lang.word2index.get(word, UNK_token)
for word in process_sentence(sentence)
]