VisionPorComputadoras

Introduction tutorial to getting started in Deep Learning with Keras

• Jorge Sanchez
• Pablo Pastore

---

Dependencies

• Python 3+
• pip >= 9.0
• virtualenv (optional)
• virtualenvwrapper (optional)
• tensorflow

Virtualenv and Virtualenvwrapper (optional)

$ pip3 install --user virtualenv virtualenvwrapper

Then add the following lines to the bottom of you ~/.bashrc file:

# virtualenv and virtualenvwrapper settings
export VIRTUALENVWRAPPER_PYTHON=/usr/bin/python3
[[ -s "$HOME/.local/bin/virtualenvwrapper.sh" ]] && source "$HOME/.local/bin/virtualenvwrapper.sh"

Update changes:

$ source ~/.bashrc

You should see an output like this:

virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/premkproject
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/postmkproject
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/initialize
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/premkvirtualenv
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/postmkvirtualenv
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/prermvirtualenv
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/postrmvirtualenv
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/predeactivate
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/postdeactivate
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/preactivate
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/postactivate
virtualenvwrapper.user_scripts creating /home/<username>/.virtualenvs/get_env_details

Create virtualenv:

$ mkvirtualenv <name> --python=/usr/bin/python3

Additional Python libs

Install other Python libraries with: $ pip install -r requirements.txt

Tensorflow

CPU only

$ pip install tensorflow==1.5

Using the GPU:

Maybe you should add the following lines to you ~/.bashrc:

# Cuda and cudnn
export PATH=$PATH:/opt/cuda/9.0/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/cudnn/v7.0:/opt/cuda/9.0/lib64

Update changes:

$ source ~/.bashrc

The tensorflow version that you should install depends on you Cuda and Cudnn version. For example for Cuda 9.0 with Cudnn 7.0 we should install the following version: $ pip install tensorflow-gpu==1.5

---

Run Notebooks

Local machine:

$ jupyter notebook

Working on a remote server:

1. Launch notebook on the server:

$ jupyter notebook --port=<port> --no-browser --ip=0.0.0.0

2. Make a tunnel on your local machine:

$ ssh -f remote_user@remote_host -L <port>:localhost:<port> -N

3. Then go to: http://localhost:/?token=<token>