The last time Hackerfall tried to access this page, it returned a not found error. A cached version of the page is below, or clickhereto continue anyway

GitHub - silicon-valley-data-science/RNN-Tutorial: Recurrent Neural Networks - A Short TensorFlow Tutorial

Setup

Clone this repo to your local machine, and add the RNN-Tutorial directory as a system variable to your ~/.profile. Instructions given for bash shell:

git clone https://github.com/silicon-valley-data-science/RNN-Tutorial
cd RNN-Tutorial
echo "export RNN_TUTORIAL=${PWD}" >> ~/.profile
echo "export PYTHONPATH=${PWD}/src:${PYTHONPATH}" >> ~/.profile
source ~/.profile

Create a Conda environment (You will need to Install Conda first)

conda create --name tf-rnn python=3
source activate tf-rnn
cd $RNN_TUTORIAL
pip install -r requirements.txt

Install TensorFlow

If you have a NVIDIA GPU with CUDA already installed

pip install tensorflow-gpu==1.0.1

If you will be running TensorFlow on CPU only (e.g. a MacBook Pro), use the following command (if you get an error the first time you run this command read below):

pip install --upgrade\
 https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.0.1-py3-none-any.whl

Error note (if you did not get an error skip this paragraph): Depending on how you installed pip and/or conda, we've seen different outcomes. If you get an error the first time, rerunning it may incorrectly show that it installs without error. Try running with pip install --upgrade https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.0.1-py3-none-any.whl --ignore-installed. The --ignore-installed flag tells it to reinstall the package. If that still doesn't work, please open an issue, or you can try to follow the advice here.

Run unittests

We have included example unittests for the tf_train_ctc.py script

python $RNN_TUTORIAL/src/tests/train_framework/tf_train_ctc_test.py

Run RNN training

All configurations for the RNN training script can be found in $RNN_TUTORIAL/configs/neural_network.ini

python $RNN_TUTORIAL/src/train_framework/tf_train_ctc.py

NOTE: If you have a GPU available, the code will run faster if you set tf_device = /gpu:0 in configs/neural_network.ini

TensorBoard configuration

To visualize your results via tensorboard:

tensorboard --logdir=$RNN_TUTORIAL/models/nn/debug_models/summary/

Add data

We have included example data from the LibriVox corpus in data/raw/librivox/LibriSpeech/. The data is separated into folders:

- Train: train-clean-100-wav (5 examples)
- Test: test-clean-wav (2 examples)
- Dev: dev-clean-wav (2 examples)

If you would like to train a performant model, you can add additional wave and txt files to these folders, or create a new folder and update configs/neural_network.ini with the folder locations

Remove additions

We made a few additions to your .profile -- remove those additions if you want, or if you want to keep the system variables, add it to your .bash_profile by running:

echo "source ~/.profile" >> .bash_profile

Next steps

We hope that our provided repo is a useful resource for getting started. Please share your experiences with adopting RNNs by contacting us or putting in pull requests for suggested changes. To stay in touch, sign up for our newsletter.

Continue reading on github.com