- Update to tensorflow 1.0.0
- Add new tutorial for TensorBoard
- Use a local virtualenv instead of polluting home directory
- Reworked installation instructions
The tutorials run through [Jupyter Notebooks][juypter].
Using git, clone this tutorial and enter that directory.
git clone https://github.com/xtoddx/tf-tutorial.git
cd tf-tutorial
Make sure your base system dependencies for pip and virtualenv, which are common python tools, are up to date. Other packages are installed into a local directory via virtualenv, so they don't need to be managed manually.
sudo easy_install --upgrade pip
sudo pip install --upgrade virtualenv
Now, create a virtual environment.
virtualenv --system-site-packages .venv
Activate the new virtual environment. This lets you use an isolated python version and dependency graph.
source .venv/bin/activate
TODO(todd): move these into requirements.txt
pip install --upgrade protobuf
TF_ROOT=https://storage.googleapis.com/tensorflow/mac/cpu/
TF_WHEEL=tensorflow-1.0.0-py2-none-any.whl
pip install --upgrade ${TF_ROOT}${TF_WHEEL}
pip install --upgrade jupyter
pip install --upgrade Pillow
Matplotlib on OSX is a bit tricky.
pip install --upgrade matplotlib
You may need to symlink /usr/local/opt/freetype/include/freetype2 to /usr/local/include/freetype
brew install freetype
ln -s /usr/local/opt/freetype/include/freetype2 /usr/local/etc/include/freetype
Launch the server. It will automatically launch a new browser window.
jupyter notebook
For any project in the notebook, you can step through the blocks using Shift+Enter to execute. The results are added to the page.