Bucket-based Tensorflow seq2seq model
This is a demo of Bucket-based Tensorflow seq2seq model, with supporting for both training and decoding.
Please see test_predict in test_bucketmodel.py for usage.
Internally, it uses tf.contrib.legacy_seq2seq.model_with_buckets to build multiple models for each bucket (all share the same set of parameters). It can handle data pre-process and data post-process for using bucket-based models.
The test can be runned with pytest after you installed pytest package (pip install pytest).
It has been tested under Tensorflow docker image, with TF version 1.0.1