def make_prediction(
self,
X_test,
remove_all_o=True,
min_conf=0.0): #not used when this is component classifier
if self.iter < 1: return None
latest_model_name = self.model_name + str(self.iter - 1)
self.tagger.open(self.model_path + latest_model_name + '.crfsuite')
y_preds = []
y_preds_mar_probs = []
for seq_idx, xseq in enumerate(X_test):
y_pred_ori = self.tagger.tag(xseq)
y_pred_mar_prob = []
for t, y in enumerate(y_pred_ori):
y_pred_mar_prob.append(self.tagger.marginal(y, t))
if st.FIXED_MIN_SEQ_PROB == None:
y_pred_post, _1, y_pred_mar_prob = utils.post_process_y(
y_pred_ori, y_pred_mar_prob, remove_all_o, min_conf)
else:
seq_prob = self.tagger.probability(y_pred_ori)
y_pred_post, _1, y_pred_mar_prob = utils.post_process_y(
y_pred_ori,
y_pred_mar_prob,
remove_all_o,
min_conf,
seq_prob=seq_prob)
y_preds.append(y_pred_post)
y_preds_mar_probs.append(y_pred_mar_prob)
self.tagger.close()
return y_preds, y_preds_mar_probs
def _vote_predictions(self, component_preds, component_marginal_prob_dists,
remove_all_o, min_conf): #need to be modified
if st.VOTE_ON_DIST is True:
vote_function = self._vote_on_dist
else:
vote_function = self._vote
voted_preds = []
pred_scores = []
for seq_idx in range(len(component_preds[0])):
voted_seq = []
score_seq = []
for bio_idx in range(len(component_preds[0][seq_idx])):
pred_4_one = [
component_preds[cp_idx][seq_idx][bio_idx]
for cp_idx in range(len(component_preds))
]
mar_prob_dist_4_one = None
if st.VOTE_ON_DIST is True:
mar_prob_dist_4_one = [
component_marginal_prob_dists[cp_idx][seq_idx][bio_idx]
for cp_idx in range(len(component_preds))
]
result, score = vote_function(pred_4_one, mar_prob_dist_4_one)
voted_seq.append(result)
score_seq.append(score)
if len(score_seq) != 0:
voted_seq, _, score_seq = utils.post_process_y(
voted_seq, score_seq, remove_all_o, min_conf)
voted_preds.append(voted_seq)
pred_scores.append(score_seq)
return voted_preds, pred_scores
def make_prediction(
self,
X_test,
remove_all_o=True,
min_conf=0.0,
link_pos=None): #not used when this is component classifier
if self.iter < 1: return None
latest_model_name = self.model_name + str(self.iter - 1)
#print('open : '+latest_model_name)
self.tagger.open(self.model_path + latest_model_name + '.crfsuite')
y_preds = []
y_preds_mar_probs = []
for seq_idx, xseq in enumerate(X_test):
y_pred_ori = self.tagger.tag(xseq) # ????????????
#if self.tagger.probability(y_pred_ori) < min_conf :
# y_preds.append([])
# y_preds_mar_probs.append([])
# continue
y_pred_mar_prob = []
for t, y in enumerate(y_pred_ori):
y_pred_mar_prob.append(self.tagger.marginal(y, t))
if st.FIXED_MIN_SEQ_PROB == None:
y_pred_post, _1, y_pred_mar_prob = utils.post_process_y(
y_pred_ori, y_pred_mar_prob, remove_all_o, min_conf,
None if link_pos == None else link_pos[seq_idx])
else:
seq_prob = self.tagger.probability(y_pred_ori)
y_pred_post, _1, y_pred_mar_prob = utils.post_process_y(
y_pred_ori,
y_pred_mar_prob,
remove_all_o,
min_conf,
None if link_pos == None else link_pos[seq_idx],
seq_prob=seq_prob)
#y_pred_mar_prob = []
#if len(y_pred_post) >0 :
# for t, y in enumerate(y_pred_ori) :
# y_pred_mar_prob.append(self.tagger.marginal(y, t))
y_preds.append(y_pred_post)
y_preds_mar_probs.append(y_pred_mar_prob)
#y_pred = [self.tagger.tag(xseq) for xseq in X_test]
self.tagger.close()
#return y_pred
return y_preds, y_preds_mar_probs
def _vote_predictions(self, component_preds, component_marginal_prob_dists,
remove_all_o, min_conf, link_pos, replace_o,
mul_ne_cnt): #need to be modified
vote_function = self._vote_on_dist
voted_preds = []
pred_scores = []
for seq_idx in range(len(component_preds[0])):
voted_seq = []
score_seq = []
for bio_idx in range(len(component_preds[0][seq_idx])):
pred_4_one = [
component_preds[cp_idx][seq_idx][bio_idx]
for cp_idx in range(len(component_preds))
]
mar_prob_dist_4_one = [
component_marginal_prob_dists[cp_idx][seq_idx][bio_idx]
for cp_idx in range(len(component_preds))
]
#for cp_idx in range(len(component_preds)):
# pred_4_one.append(component_preds[cp_idx][seq_idx][bio_idx])
# mar_prob_4_one.append()
result, score = vote_function(pred_4_one, mar_prob_dist_4_one)
voted_seq.append(result)
score_seq.append(score)
if len(score_seq) != 0:
voted_seq, _, score_seq = utils.post_process_y(
voted_seq,
score_seq,
remove_all_o,
min_conf,
None if link_pos == None else link_pos[seq_idx],
replace_o=replace_o,
mul_ne_cnt=mul_ne_cnt)
voted_preds.append(voted_seq)
pred_scores.append(score_seq)
return voted_preds, pred_scores