Skip to the content.

This is an old version of the code that is not as effective. It is included for completeness.

Compile

  1. Install DyNet for C++ as described here.
  2. Run make in this directory.

Train and run

This command will run the system assuming you have compiled it and are now in the root directory of this repository:

./src/old-cpp-version/predictor
  --dynet-mem 3048
  --dynet-autobatch 1
  --dynet-weight-decay 0.00000001
  --data-train data/list.ubuntu.train.txt
  --data-dev data/list.ubuntu.dev.txt
  --data-eval data/list.ubuntu.test.txt
  --log-freq 5000
  --dev-freq 1
  --trainer sgd
  --dim-input 100
  --model ff
  --prefix example-run-cpp
  --instance-type selection
  --learning-rate 0.1
  --selection-proportion 0.1
  --input-hand-crafted
  --nonlinearity tanh
  --input-text
  --dim-ff-hidden 64
  --layers-ff-pair 2
  --max-link-length 100
  --max-iterations 200
  --word-vector-init ubuntu-all.tok-user.unked-100.w2v.txt
  --no-improvement-cutoff 30
  --clipping-threshold 1.0