def evaluate_scws(sess, en_dh, batch_size, en_ss, en_srl):
global global_maxSimC
global global_avgSimC
[senseVec, senseScore] = getEmbedding(sess,
en_dh,
batch_size,
en_ss,
en_srl.embedded_sense_input,
en_srl.selected_sense_input_indices,
scws=True)
indices1 = range(0, len(senseVec), 2)
indices2 = range(1, len(senseVec), 2)
avgSimC = calAvgSimC(en_dh.scws_test_score, senseVec[indices1],
senseScore[indices1], senseVec[indices2],
senseScore[indices2])
maxSimC = calMaxSimC(en_dh.scws_test_score, senseVec[indices1],
senseScore[indices1], senseVec[indices2],
senseScore[indices2])
scores = [maxSimC, avgSimC]
score_name = ['MaxSimC', 'AvgSimC']
print 'scws AvgSimC =', '{:.5f}'.format(
avgSimC), 'MaxSimC =', '{:.5f}'.format(maxSimC)
ret_str = 'scws AvgSimC = {:.5f}'.format(
avgSimC) + ' MaxSimC = {:.5f}'.format(maxSimC)
return ret_str, scores, score_name
def evaluate_scws(sess, en_dh, batch_size, en_ss, en_srl):
global global_maxSimC
global global_avgSimC
[senseVec, senseScore] = getEmbedding(sess,
en_dh,
batch_size,
en_ss,
en_srl.embedded_sense_input,
en_srl.selected_sense_input_indices,
scws=True)
indices1 = range(0, len(senseVec), 2)
indices2 = range(1, len(senseVec), 2)
avgSimC = calAvgSimC(en_dh.scws_test_score, senseVec[indices1],
senseScore[indices1], senseVec[indices2],
senseScore[indices2])
maxSimC = calMaxSimC(en_dh.scws_test_score, senseVec[indices1],
senseScore[indices1], senseVec[indices2],
senseScore[indices2])
scores = [maxSimC, avgSimC]
if maxSimC > ckpt_threshold and maxSimC > global_maxSimC:
global_maxSimC = maxSimC
eval_save_model(sess, '-bestMaxC-')
if avgSimC > ckpt_threshold and avgSimC > global_avgSimC:
global_avgSimC = avgSimC
eval_save_model(sess, '-bestAvgC-')
print 'scws AvgSimC =', '{:.5f}'.format(
avgSimC), 'MaxSimC =', '{:.5f}'.format(maxSimC)
ret_str = 'scws AvgSimC = {:.5f}'.format(
avgSimC) + ' MaxSimC = {:.5f}'.format(maxSimC)
return ret_str
def evaluate_bcws(sess, en_dh, ch_dh, batch_size, en_ss, ch_ss, en_opt,
ch_opt):
global global_bmaxSimC
global global_bavgSimC
[en_senseVec, en_senseScore] = getEmbedding(sess, en_dh, batch_size, en_ss,
*bcws_dict[en_opt])
[ch_senseVec, ch_senseScore] = getEmbedding(sess, ch_dh, batch_size, ch_ss,
*bcws_dict[ch_opt])
assert en_dh.test_score == ch_dh.test_score
avgSimC = calAvgSimC(en_dh.test_score, en_senseVec, en_senseScore,
ch_senseVec, ch_senseScore)
maxSimC = calMaxSimC(en_dh.test_score, en_senseVec, en_senseScore,
ch_senseVec, ch_senseScore)
if maxSimC > bckpt_threshold and maxSimC > global_bmaxSimC:
global_bmaxSimC = maxSimC
eval_save_model(sess, '-bestbMaxC_{}_{}-'.format(en_opt, ch_opt))
if avgSimC > bckpt_threshold and avgSimC > global_bavgSimC:
global_bavgSimC = avgSimC
eval_save_model(sess, '-bestbAvgC_{}_{}-'.format(en_opt, ch_opt))
ret_str = 'bcws {} {} AvgSimC = '.format(en_opt, ch_opt) + '{:.5f}'.format(
avgSimC) + ' MaxSimC = ' + '{:.5f}'.format(maxSimC)
print(ret_str)
return ret_str
def evaluate_bcws(sess, en_dh, ch_dh, batch_size, en_ss, ch_ss, en_srl, ch_srl):
global global_bmaxSimC
global global_bavgSimC
[en_senseVec, en_senseScore] = getEmbedding(sess, en_dh, batch_size, en_ss, en_srl.embedded_sense_input, en_srl.selected_sense_input_indices)
[ch_senseVec, ch_senseScore] = getEmbedding(sess, ch_dh, batch_size, ch_ss, ch_srl.embedded_sense_input, ch_srl.selected_sense_input_indices)
assert en_dh.test_score == ch_dh.test_score
avgSimC = calAvgSimC(en_dh.test_score, en_senseVec, en_senseScore, ch_senseVec, ch_senseScore)
maxSimC = calMaxSimC(en_dh.test_score, en_senseVec, en_senseScore, ch_senseVec, ch_senseScore)
print ('bcws AvgSimC =','{:.5f}'.format(avgSimC), 'MaxSimC =','{:.5f}'.format(maxSimC))
def evaluate(sess):
with Timer("get embedding for 1st half of evaluation data"):
[senseVec, senseScore] = getEmbedding(sess,testData, testLocation)
indices1 = range(0,len(senseVec),2)
indices2 = range(1,len(senseVec),2)
avgSimC = calAvgSimC(test_score, senseVec[indices1], senseScore[indices1], senseVec[indices2], senseScore[indices2])
maxSimC = calMaxSimC(test_score, senseVec[indices1], senseScore[indices1], senseVec[indices2], senseScore[indices2])
scores = [maxSimC, avgSimC]
score_name = ['MaxSimC', 'AvgSimC']
print 'AvgSimC =','{:.5f}'.format(avgSimC), 'MaxSimC =','{:.5f}'.format(maxSimC)
return [scores, score_name]