/ltr-dnn

LTR DNN in tensorflow, an improvement of DSSM

Primary LanguagePythonMIT LicenseMIT

LTR-DNN

A tensorflow implementation of LTR DNN semantic similarity model.

Tensorflow v1.2 on CPU is used for testing.

Model Structure:

Model Structure

Example:

# train
python -u train.py \
--train_bs=128 \
--max_epoch=25 \
--eps=0.2 \
--eval_steps=20 \
--max_iter=1000 \
--embedding_file=./data/words_embedding \
--train_file=./data/data_train_example.tsv \
--test_file=./data/data_test_example.tsv \
--valid_file=./data/data_test_example.tsv

# predict
python -u predict.py \
--predict_file=./data/data_predict_example.tsv

Reference:

  • It's an LTR improvement of DSSM