No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Wanjia He Wanjia He
Wanjia He and Wanjia He git readme
Latest commit fca3391 Apr 21, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Data git readme Apr 21, 2017
Scripts first commit Jan 17, 2017
README.md Update README.md Jan 20, 2017

README.md

Multi-view-neural-acoustic-words-embeddings

Prerequisites

Python --- 2.7.10, Tensorflow --- 0.9.0rc0, numpy, scipy

How to use

To help organizing the code, here is my working directory

/Multi-view_embedding
    /Data
    /Outputs
    /Saved_Models
    /Scripts

To train a model with 2-layer bidirectional LSTMs on both views:

python 2biLSTM.py <parameters>
available parameters include:
  -lr: learning rate, 0.001 by default (for LSTMs with more than 1 layer, 0.0001 is recommended)
  -obj: objectives, 0(obj0) by default, 02 = averaged sum of obj0 and obj2, must be a subset of 0 1 2 3
  -hs: hidden size, 512 by default
  -m: margin, 0.5 by default
  -kp: keep probability of dropout, 0.6 by default
  -bs: batch size, 20 by default
  -epo: epochs, 20 by default
  -inits: initialization scale of LSTM cells, 0.05 by default
to train our best model:
python 2biLSTM.py -lr 0.0001 -obj 02 -m 0.4 -epo 1000