The result is bad

[duguiming111]

I use the zh-en corpus of iwslt2015, and the Bleu value can only reach 5.3. I find that there are many such problems in the issues, which can't reach the result you give. Is there any problem in our operation?

[duguiming111]

When I run the code, python3 nmt.py --attention=scaled_luong --src=zh --tgt=en --vocab_prefix=tmp/nmt_data_zh_en/vocab --train_prefix=tmp/nmt_data_zh_en/train --dev_prefix=tmp/nmt_data_zh_en/tst2012 --test_prefix=tmp/nmt_data_zh_en/tst2013 --out_dir=tmp/nmt_attention_model --num_train_steps=12000 --steps_per_stats=100 --num_layers=2 --num_units=512 --encoder_type=bi --dropout=0.2 --decay_scheme=luong234 --metrics=bleu The best value of blue is 11.4，are there sone problems?

[maharajbrahma]

Bleu score depends on various factors like architecture used, hyper parameters, number of sentences in data set.

[watermelona]

@duguiming111 i am the begginer of the nmt, i can not find the zh-en corpus of iwslt2015, can you tell me how to get this dataset and vocab ?