def main() -> None:
(x_train, t_train), (x_test, t_test) = sequence.load_data('addition.txt')
x_train, x_test = x_train[:, ::-1], x_test[:, ::-1]
char_to_id, id_to_char = sequence.get_vocab()
vocab_size = len(char_to_id)
wordvec_size = 16
hidden_size = 128
batch_size = 128
max_epoch = 25
max_grad = 5.0
model = PeekySeq2seq(vocab_size, wordvec_size, hidden_size)
optimizer = Adam()
trainer = Trainer(model, optimizer)
acc_list = []
for epoch in range(1, max_epoch+1):
trainer.fit(x_train, t_train, max_epoch=1, batch_size=batch_size, max_grad=max_grad)
correct_num = 0
for i in range(len(x_test)):
question, correct = x_test[[i]], t_test[[i]]
verbose = i < 10
correct_num += eval_seq2seq(model, question, correct, id_to_char, verbose)
acc = float(correct_num) / len(x_test)
acc_list.append(acc)
print(f'val acc {acc*100}%')
print('DONE')
def train_eval(x_train, x_test, is_peeky):
if is_peeky:
model = PeekySeq2seq(vocab_size, wordvec_size, hidden_size)
else:
model = Seq2seq(vocab_size, wordvec_size, hidden_size)
optimizer = Adam()
trainer = Trainer(model, optimizer)
acc_list = []
for epoch in range(max_epoch):
trainer.fit(x_train,
t_train,
max_epoch=1,
batch_size=batch_size,
max_grad=max_grad)
correct_num = 0
for i in range(len(x_test)):
question, correct = x_test[[i]], t_test[[i]]
verbose = i < 10
correct_num += eval_seq2seq(model, question, correct, id_to_char,
verbose)
acc = float(correct_num) / len(x_test)
acc_list.append(acc)
print('val acc %.3f%%' % (acc * 100))
return acc_list
def main():
# データセットの読み込み
(x_train, t_train), (x_test, t_test) = sequence.load_data('addition.txt')
char_to_id, id_to_char = sequence.get_vocab()
# 入力列を逆順にするとSeq2Se2の精度が上がるらしいが。。。クソ理論
is_reverse = True
if is_reverse:
x_train, x_test = x_train[:, ::-1], x_test[:, ::-1]
# ハイパーパラメータの設定
vocab_size = len(char_to_id)
wordvec_size = 16
hidden_size = 128
batch_size = 128
max_epoch = 25
max_grad = 5.0
# モデル/オプティマイザ/トレーナーの生成
# model = Seq2seq(vocab_size, wordvec_size, hidden_size)
model = PeekySeq2seq(vocab_size, wordvec_size, hidden_size)
optimizer = Adam()
trainer = Trainer(model, optimizer)
acc_list = []
for epoch in range(max_epoch):
trainer.fit(x_train,
t_train,
max_epoch=1,
batch_size=batch_size,
max_grad=max_grad)
correct_num = 0
for i in range(len(x_test)):
question, correct = x_test[[i]], t_test[[i]]
verbose = i < 10
correct_num += eval_seq2seq(model, question, correct, id_to_char,
verbose)
acc = float(correct_num) / len(x_test)
acc_list.append(acc)
print(f'val acc {acc * 100}')
xs = corpus[:-1]
ts = corpus[1:]
# ハイパーパラメータの設定
vocab_size = len(word_to_id)
wordvec_size = 16
hidden_size = 128
batch_size = 1
max_epoch = 50
max_grad = 5.0
sample_size = 100
lr = 0.001
time_size = 35
#モデルの生成
model = PeekySeq2seq(vocab_size, wordvec_size, hidden_size)
optimizer = Adam()
trainer = RnnlmTrainer(model, optimizer)
#学習
best_ppl = float('inf')
t1 = time.time()
for epoch in range(max_epoch):
trainer.fit(xs, ts, max_epoch=1, batch_size=batch_size, max_grad=max_grad)
model.reset_state()
ppl = eval_perplexity(model, corpus)
print('valid perplexity: ', ppl)
if best_ppl > ppl:
best_ppl = ppl