def search_hparams():
verbose=False
ngrams = [{'ngram':x} for x in range(1,8,1)]
smooth = [{'est':'mle'}, {'est':'laplace'}, {'est':'good-turing'}]
smooth.extend([{'est':'add-delta', 'delta':x/10} for x in range(1,10,1)])
with open('hmm_res.csv', 'w') as f:
for config in itertools.product(ngrams, smooth):
config = {**config[0], **config[1]}
if 'conf_mat' in locals():
del conf_mat
for iters in range(7):
hmm = HMM(config)
if 'conf_mat' in locals():
conf_mat += hmm.prep_data().shuffle(None).split(0.1).train().eval()
else:
conf_mat = hmm.prep_data().shuffle(None).split(0.1).train().eval()
res_str = '{};'.format(config)
print("Configuration = {}: ".format(config))
precision, recall = metrics.MicroAvg(conf_mat)
f1 = metrics.Fscore(precision, recall, 1)
res_str += '{};'.format(f1)
print('MicroAvg:',precision,recall,f1)
precision, recall = metrics.MacroAvg(conf_mat)
f1 = metrics.Fscore(recall, precision, 1)
res_str += '{};'.format(f1)
print('MacroAvg:', precision, recall, f1)
acc = metrics.AvgAcc(conf_mat)
res_str += '{};'.format(acc)
print('AvgAcc:', acc)
f.write(res_str+'\n')
conf_mat = metrics.NormalizeConfusion(conf_mat)
if verbose:
print('ConfMat:\n', np.array_str(conf_mat, max_line_width=300, precision=4))
print('----------------------------------------------')
def search_hparams():
verbose = False
with open('crf_word_res.csv','w') as f:
for num_ftrs in range(len(CRF.WORD_FTRS)):
num_ftrs += 1
for ftrs in combinations(CRF.WORD_FTRS, num_ftrs):
config = {'ftrs':ftrs}
if 'conf_mat' in locals():
del conf_mat
for i in range(7):
crf = CRF(config)
if 'conf_mat' in locals():
conf_mat += crf.prep_data().shuffle(None).split(0.1).train().eval()
else:
conf_mat = crf.prep_data().shuffle(None).split(0.1).train().eval()
res_str = '{};'.format(config)
print("Configuration = {}: ".format(config))
precision, recall = metrics.MicroAvg(conf_mat)
f1 = metrics.Fscore(precision, recall, 1)
res_str += '{};'.format(f1)
print('MicroAvg:',precision,recall,f1)
precision, recall = metrics.MacroAvg(conf_mat)
f1 = metrics.Fscore(recall, precision, 1)
res_str += '{};'.format(f1)
print('MacroAvg:', precision, recall, f1)
acc = metrics.AvgAcc(conf_mat)
res_str += '{};'.format(acc)
print('AvgAcc:',acc)
f.write(res_str+'\n')
conf_mat = metrics.NormalizeConfusion(conf_mat)
if verbose:
print('ConfMat:\n', np.array_str(conf_mat, max_line_width=300, precision=4))
print('----------------------------------------------')
def search_hparams():
verbose = False
with open('crf_sentence_res.csv', 'w') as f:
poss_words_ids = [[0], [-1, 0, 1], [-2, 0, 2], [-2, -1, 0, 1, 2]]
for poss_words_id in poss_words_ids:
config = {'extr_word_idx': poss_words_id}
for num_word_ftrs in range(6, len(CRF.WORD_FTRS)):
num_word_ftrs += 1
for word_ftrs in combinations(CRF.WORD_FTRS, num_word_ftrs):
config['word_ftrs'] = word_ftrs
for num_stc_ftrs in range(len(CRF.STC_FTRS) + 1):
for stc_ftrs in combinations(CRF.STC_FTRS,
num_stc_ftrs):
config['stc_ftrs'] = stc_ftrs
if 'conf_mat' in locals():
del conf_mat
for i in range(7):
#print(config)
crf = CRF(config)
if 'conf_mat' in locals():
conf_mat += crf.prep_data().shuffle(
0).split(0.1).train().eval()
else:
conf_mat = crf.prep_data().shuffle(
0).split(0.1).train().eval()
res_str = '{};'.format(config)
print("Configuration = {}: ".format(config))
precision, recall = metrics.MicroAvg(conf_mat)
f1 = metrics.Fscore(precision, recall, 1)
res_str += '{};'.format(f1)
print('MicroAvg:', precision, recall, f1)
precision, recall = metrics.MacroAvg(conf_mat)
f1 = metrics.Fscore(recall, precision, 1)
res_str += '{};'.format(f1)
print('MacroAvg:', precision, recall, f1)
acc = metrics.AvgAcc(conf_mat)
res_str += '{};'.format(acc)
print('AvgAcc:', acc)
f.write(res_str + '\n')
conf_mat = metrics.NormalizeConfusion(conf_mat)
if verbose:
print(
'ConfMat:\n',
np.array_str(conf_mat,
max_line_width=300,
precision=4))
print(
'----------------------------------------------'
)
def search_hparams():
verbose = False
with open('rnn_res.csv', 'a') as f:
num_layers = [1, 2, 3, 4, 5, 6]
n_hidden = [32, 64, 128]
emb = ['one-hot', 'mds', 'nn']
win_len = [4]
for config_tuple in product(num_layers, n_hidden, emb, win_len):
config = {
'n_layers': config_tuple[0],
'hidden_dim': config_tuple[1],
'embedding': config_tuple[2],
'win_len': config_tuple[3]
}
if 'conf_mat' in locals():
del conf_mat
for i in range(5):
rnn = Encoder(config)
if 'conf_mat' in locals():
conf_mat += rnn.prep_model().shuffle(None).split(
0.05).train(epochs=10000, lr=1e-2, alg='adamw').eval()
else:
conf_mat = rnn.prep_model().shuffle(None).split(
0.05).train(epochs=10000, lr=1e-2, alg='adamw').eval()
res_str = '{};'.format(config)
print("Configuration = {}: ".format(config))
precision, recall = metrics.MicroAvg(conf_mat)
f1 = metrics.Fscore(precision, recall, 1)
res_str += '{};'.format(f1)
print('MicroAvg:', precision, recall, f1)
precision, recall = metrics.MacroAvg(conf_mat)
f1 = metrics.Fscore(recall, precision, 1)
res_str += '{};'.format(f1)
print('MacroAvg:', precision, recall, f1)
acc = metrics.AvgAcc(conf_mat)
res_str += '{};'.format(acc)
print('AvgAcc:', acc)
f.write(res_str + '\n')
conf_mat = metrics.NormalizeConfusion(conf_mat)
if verbose:
print('ConfMat:\n',
np.array_str(conf_mat, max_line_width=300, precision=4))
print('----------------------------------------------')
f.flush()