def augment(inputs, outputs, tags, hallucinator, min_len=3, max_len=10):
temp = [(''.join(inp), ''.join(out)) for inp, out in zip(inputs, outputs)]
# aligned returns pairs of strings with spaces for null alignments
aligned = align.Aligner(temp).alignedpairs
new_inputs = []
new_outputs = []
new_tags = []
for k, (src, trg) in enumerate(aligned):
good_ranges = find_good_range(src, trg, min_len, max_len)
if good_ranges:
new_src, new_trg = list(src), list(trg)
for good_range in good_ranges:
s, e = good_range
gold_seq = new_src[s: e]
hallucinated_seq = hallucinator.sample(gold_seq)
new_src[s: e] = new_trg[s: e] = hallucinated_seq
# trim, unless src and trg have an aligned whitespace
new_i1 = [c for i, c in enumerate(new_src)
if (c.strip() or (new_src[i] == new_trg[i] == ' '))]
new_o1 = [c for i, c in enumerate(new_trg)
if (c.strip() or (new_src[i] == new_trg[i] == ' '))]
new_inputs.append(new_i1)
new_outputs.append(new_o1)
new_tags.append(tags[k])
return new_inputs, new_outputs, new_tags
def smart_align(pairs, align_symbol=ALIGN_SYMBOL,
iterations=150, burnin=5, lag=1, mode='crp', **kwargs):
return align.Aligner(pairs,
align_symbol=align_symbol,
iterations=iterations,
burnin=burnin,
lag=lag,
mode=mode).alignedpairs
def augment(inputs, outputs, tags, characters):
temp = [(''.join(inputs[i]), ''.join(outputs[i]))
for i in range(len(outputs))]
aligned = align.Aligner(temp, align_symbol=' ').alignedpairs
vocab = list(characters)
try:
vocab.remove(u" ")
except:
pass
new_inputs = []
new_outputs = []
new_tags = []
for k, item in enumerate(aligned):
#print(''.join(inputs[k]) + '\t' + ''.join(outputs[k]))
i, o = item[0], item[1]
good_range = find_good_range(i, o)
#print(good_range)
if good_range:
new_i, new_o = list(i), list(o)
for r in good_range:
s = r[0]
e = r[1]
if (e - s > 5): #arbitrary value
s += 1
e -= 1
for j in range(s, e):
if random() > 0.5: #arbitrary value
nc = choice(vocab)
new_i[j] = nc
new_o[j] = nc
new_i1 = [
c for l, c in enumerate(new_i)
if (c.strip() or (new_o[l] == ' ' and new_i[l] == ' '))
]
new_o1 = [
c for l, c in enumerate(new_o)
if (c.strip() or (new_i[l] == ' ' and new_o[l] == ' '))
]
new_inputs.append(new_i1)
new_outputs.append(new_o1)
new_tags.append(tags[k])
else:
new_inputs.append([])
new_outputs.append([])
new_tags.append([])
return new_inputs, new_outputs, new_tags
def process(organism_ids, align_method, similarity_mode,
power_alpha=cs.ALPHA_BIAS, check=True, visual=False):
# load bio_net object
bio_net = initialize_network(organism_ids, align_method,
similarity_mode, power_alpha)
# create aligner object
aligner = align.Aligner(align_method)
alignment = aligner.align(bio_net, check=check)
# visualization
if visual:
visualize.gephi_organism_ppi(bio_net.org1)
visualize.gephi_organism_ppi(bio_net.org2)
visualize.gephi_network_aligned(alignment, bio_net)
visualize.gephi_network_aligned_comp(alignment, bio_net)
return alignment
def med_align(wordpairs, align_symbol):
a = align.Aligner(wordpairs, align_symbol=align_symbol, mode='med')
return a.alignedpairs
def mcmc_align(wordpairs, align_symbol):
a = align.Aligner(wordpairs, align_symbol=align_symbol)
return a.alignedpairs
citationforms = {
c
for c in citationforms if citationforms[c] > 4 and citationforms[c] /
float(citationforms[c] + negcitationforms[c]) >= 0.95
}
for l in lines2:
msd1, form1, msd2, form2 = l.split(u'\t')
if msd1 in citationforms and msd1 != msd2:
traindata1.append((form1, form2, msd2))
if msd2 in citationforms and msd1 != msd2:
traindata1.append((form2, form1, msd1))
if task == 1 or not constrained:
wordpairs = [(x[0], x[1]) for x in traindata1]
a = align.Aligner(wordpairs, align_symbol=u'_', iterations=30)
traindata1 = [(traindata1[i][0], traindata1[i][1], traindata1[i][2],
a.alignedpairs[i][0], a.alignedpairs[i][1])
for i in range(len(traindata1))]
C, V = consvowOCP.candv(words)
# Lemma > form
if task == 1 or not constrained:
for lemma, form, msd, lemmaaligned, formaligned in traindata1:
if msd not in fromlemma:
fromlemma[msd] = []
if msd not in tolemma:
tolemma[msd] = []
alignedpair1 = (lemmaaligned, formaligned)
def mcmc_align(wordpairs, align_symbol):
a = align.Aligner(wordpairs, align_symbol=align_symbol, random_seed=42)
return a.alignedpairs
