Machine Learning Cookbook

Checkout Machine Learning In Hindsight

1. Install Python and Virtualenv
see: https://www.tensorflow.org/install/
Use virtualenv to created walled off python envrionment
MACOSX

    $ brew install python python3
    # next use the python package manager 'pip' to install virtualenv
    $ sudo easy_install pip
    $ sudo easy_install pip3
    $ pip3 install --upgrade pip
    $ pip3 install --user virtualenv
    $ pip3 install --user virtualenvwrapper
    $ pip3 install numpy scipy pillow
    $ pip3 install matplotlib
    $ pip3 install future
    $ mkdir -p ~/virtualenvs

Ubuntu

    $ sudo apt-get install python-pip python3-pip
    $ sudo apt-get install build-essential libssl-dev libffi-dev python3-dev
    $ sudo apt-get install libfreetype6-dev libxft-dev
    $ pip3 install --upgrade pip
    # next use the python package manager 'pip' to install virtualenv
    $ pip3 install --user virtualenv
    $ pip3 install --user virtualenvwrapper
    # add virtualenv to your path
    $ pip3 install numpy
    $ pip3 install pillow
    $ pip3 install future
    $ sudo pip3 install matplotlib
    $ sudo apt-get install virtualenv
    $ mkdir -p ~/virtualenvs

2. Setup your paths
MAC OSX

    # WARNING: you may have to set PYTHONPATH to include ~/library/Python/2.7/lib/python/site-packages/
    # Place the following in your .profile settings
    VIRTUALENVWRAPPER_PYTHON=$(which python3)
    export VIRTUALENVWRAPPER_PYTHON
    export WORKON_HOME=~/virtualenvs
    if [ -f "/usr/local/bin/virtualenvwrapper.sh" ]; then
        . /usr/local/bin/virtualenvwrapper.sh
    fi
    PATH="$PATH:$HOME/Library/Python/2.7/bin"
    export PATH
    # end .profile changes

Ubuntu

    a. In your $HOME/.profile-extra,
    PATH="$HOME/.local/bin:$PATH"
    export PATH
    VIRTUALENVWRAPPER_PYTHON=$(which python3)
    export VIRTUALENVWRAPPER_PYTHON
    export WORKON_HOME=~/virtualenvs
    if [ -f "$HOME/.local/bin/virtualenvwrapper.sh" ]; then
        . "$HOME/.local/bin/virtualenvwrapper.sh"
    elif [ -f "/usr/local/bin/virtualenvwrapper.sh" ]; then
        . /usr/local/bin/virtualenvwrapper.sh
    fi
    # end .profile changes
    b. In your $HOME/.bashrc-extra,
    . "$HOME/.local/bin/virtualenvwrapper.sh"

    # end .profile changes

3. Create working directory

MACOSX

    $ mkdir -p ~/work/Foo/project1
    $ cd ~/work/Foo
    $ virtualenv -p /usr/local/bin/python3 project1

Ubuntu

    $ mkdir -p ~/work/Foo/project1
    $ cd ~/work/Foo
    $ virtualenv -p /usr/bin/python3.5 project1

4. Install TensorFlow:
NOTE: Use python3 instead of python 2

MACOSX & Ubuntu

 
    $ mkdir -p ~/tensorflow3/src
    # remove the  "-p python3" from the command line if you want a python2 package installation
    $ virtualenv --system-site-packages -p python3 ~/tensorflow3
    $ . ~/tensorflow3/bin/activate
    (tensorflow3) $ cd ~/tensorflow3
    (tensorflow3) $ easy_install -U pip
    # for Ubuntu, upgrade tensorflow-gpu or tensorflow. For Mac, the only option is tensorflow (the CPU version).
    (tensorflow3) $ pip3 install --upgrade tensorflow-gpu
    # type
    # $ deactivate
    # at the shell prompt to leave the virtual environment

5. Install Jupyter Notebook
MACOSX & Ubuntu

    $ . ~/tensorflow3/bin/activate
    (tensorflow3) $ pip3 install jupyter
    # we need to configure jupyter to be remotely accessible
    (tensorflow3) $ cd ~/tensorflow3/src

See Jupyter.org for more info on what jupyter notebook is and how to use it.

6. Optional: Make the jupyter notebook accessible remotely
By default, the jupyter notebook is accessible only via localhost. Do the
following to make it available remotely.

MACOSX & Ubuntu

    $ . ~/tensorflow3/bin/activate
    (tensorflow3) $ jupyter notebook --generate-config
    # this creates a file ~/.jupyter/jupyter_notebook_config.py 
    #
    # set the password using:
    (tensorflow3) $ jupyter notebook password
    # generate a certificate for using https
    (tensorflow3) $ openssl req -x509 -nodes -days 365 -newkey rsa:1024 -keyout jupyter.key -out jupyter.pem
    (tensorflow3) $ mv jupyter.key jupyter.pem  ~/.jupyter/
    # Edit the ~/.jupyter/jupyter_notebook_config.json file to set the additional
    # settings:
Begin File contents of: ~/.jupyter/jupyter_notebook_config.json
    {
  "NotebookApp": {
    "certfile" : "/home/{my-user-name}/.jupyter/jupyter.pem",
    "keyfile" : "/home/{my-user-name}/.jupyter/jupyter.key",
    "port": 9111,
    "open_browser": false,
    "ip": "my-hostname-or-ip-address",
    "password": "generated-sha-do-not-change"
  }
}
End File contents of: ~/.jupyter/jupyter_notebook_config.json

7. Run the Jupyter Notebook
MACOSX & Ubuntu

Notebooks

Here are some basic jupyter notebook worksheets illustrating common algorithms. Save this to your work area, ~/tensorflow3/src, and play around with them.

• TensorFlowBasicUsage.ipynb this is a basic jupyter notebook file showing linear regression
• TensorFlowMnistForMLBeginners.ipynb this shows a simple net for the MNIST digit data
• TensorFlowMnistConvoNet.ipynb this shows a multi-layer convolution net for the MNIST digit data with better than 99% accuracy.

Starting jupyter

    $ . ~/tensorflow3/bin/activate
    (tensorflow3) $ cd ~/tensorflow3/src
    (tensorflow3) $ jupyter notebook
    # open the link in your browser and you should see the various worksheets that you've saved. Click on one
    # to run.