def graph_input_prune(gr, ypred, N, skip_oneclass=False):
""" Given a gr and a given scoring, keep only top N s1 for each s0,
and stash the others away to _x-suffixed keys (for potential recovery). """
slices = []
def prune_filter(ypred, N):
""" yield (index, passed) tuples """
ys = sorted(enumerate(ypred), key=lambda yy: yy[1], reverse=True)
i = 0
for n, y in ys:
yield n, (i < N)
i += 1
# Go through (s0, s1), keeping track of the beginning of the current
# s0 block, and appending pruned versions
i = 0
grp = dict([(k, []) for k in gr.keys()] + [(k + '_x', [])
for k in gr.keys()])
for j in range(len(gr['si0']) + 1):
if j < len(gr['si0']) and (j == 0 or
np.all(gr['si0'][j] == gr['si0'][j - 1])):
# within same-s0 block, carry on
continue
# block boundary
# possibly check if we have both classes picked (for training)
if skip_oneclass:
n_picked = 0
for n, passed in prune_filter(ypred[i:j], N):
if not passed:
break
n_picked += gr['score'][i + n] > 0
if n_picked == 0:
# only false; tough luck, prune everything for this s0
for k in gr.keys():
grp[k + '_x'] += list(gr[k][i:j])
i = j
continue
# append pruned subset
for n, passed in prune_filter(ypred[i:j], N):
for k in gr.keys():
if passed:
grp[k].append(gr[k][i + n])
else:
grp[k + '_x'].append(gr[k][i + n])
i = j
return graph_nparray_anssel(grp)
def graph_input_prune(gr, ypred, N, skip_oneclass=False):
""" Given a gr and a given scoring, keep only top N s1 for each s0,
and stash the others away to _x-suffixed keys (for potential recovery). """
slices = []
def prune_filter(ypred, N):
""" yield (index, passed) tuples """
ys = sorted(enumerate(ypred), key=lambda yy: yy[1], reverse=True)
i = 0
for n, y in ys:
yield n, (i < N)
i += 1
# Go through (s0, s1), keeping track of the beginning of the current
# s0 block, and appending pruned versions
i = 0
grp = dict([(k, []) for k in gr.keys()] + [(k+'_x', []) for k in gr.keys()])
for j in range(len(gr['si0']) + 1):
if j < len(gr['si0']) and (j == 0 or np.all(gr['si0'][j] == gr['si0'][j-1])):
# within same-s0 block, carry on
continue
# block boundary
# possibly check if we have both classes picked (for training)
if skip_oneclass:
n_picked = 0
for n, passed in prune_filter(ypred[i:j], N):
if not passed:
break
n_picked += gr['score'][i + n] > 0
if n_picked == 0:
# only false; tough luck, prune everything for this s0
for k in gr.keys():
grp[k+'_x'] += list(gr[k][i:j])
i = j
continue
# append pruned subset
for n, passed in prune_filter(ypred[i:j], N):
for k in gr.keys():
if passed:
grp[k].append(gr[k][i + n])
else:
grp[k+'_x'].append(gr[k][i + n])
i = j
return graph_nparray_anssel(grp)
def graph_input_unprune(gro, grp, ypred, xval):
""" Reconstruct original graph gro from a pruned graph grp,
with predictions set to always False for the filtered out samples.
(xval denotes how the False is represented) """
if 'score_x' not in grp:
return grp, ypred # not actually pruned
gru = dict([(k, list(grp[k])) for k in gro.keys()])
# XXX: this will generate non-continuous s0 blocks,
# hopefully okay for all ev tools
for k in gro.keys():
gru[k] += grp[k+'_x']
ypred = list(ypred)
ypred += [xval for i in grp['score_x']]
ypred = np.array(ypred)
return graph_nparray_anssel(gru), ypred
def graph_input_unprune(gro, grp, ypred, xval):
""" Reconstruct original graph gro from a pruned graph grp,
with predictions set to always False for the filtered out samples.
(xval denotes how the False is represented) """
if 'score_x' not in grp:
return grp, ypred # not actually pruned
gru = dict([(k, list(grp[k])) for k in gro.keys()])
# XXX: this will generate non-continuous s0 blocks,
# hopefully okay for all ev tools
for k in gro.keys():
gru[k] += grp[k + '_x']
ypred = list(ypred)
ypred += [xval for i in grp['score_x']]
ypred = np.array(ypred)
return graph_nparray_anssel(gru), ypred
def load_set(self, fname, lists=None):
if lists:
s0, s1, y = lists
else:
# s0, s1, y = loader.load_msrpara(fname) #set it free is we decide not to use quora dataset
s0, s1, y = loader.load_quora(fname)
if self.vocab is None:
vocab = Vocabulary(s0 + s1,
prune_N=self.c['embprune'],
icase=self.c['embicase'])
else:
vocab = self.vocab
si0, sj0 = vocab.vectorize(s0, self.emb, spad=self.s0pad)
si1, sj1 = vocab.vectorize(s1, self.emb, spad=self.s1pad)
f0, f1 = nlp.sentence_flags(s0, s1, self.s0pad, self.s1pad)
gr = graph_nparray_anssel(
graph_input_anssel(si0, si1, sj0, sj1, None, None, y, f0, f1))
return (gr, y, vocab)
