Setup PyCharm for Deep learning with TensorFlow, Keras and Jupyter (with virtualenv)

This is a startup post to get your dev environment setup for diving into Deep Learning. I have chosen to begin with TensorFlow and Keras for this task. I would be using the Jupyter notebook for demonstrating the dev setup.

Note: Jupyter itself is self sufficient for this task,
but I am using PyCharm for this task just to test the interoperability of PyCharm 
with other ecosystem stack.

Create a new project with virtual env

Create a new Pure Python project in PyCharm and provide the settings for a virtual env. Pre-requisites for this step are python and virtual env. Make sure you have these steps before creation of new project:

• Install Python
• Install pip
• Install virtualenv (pip install virtualenv)

PyCharm settings for new project. Make sure you chose python 3.6 for the virtual env :

 

Activate the virtual env if not already activated:

• Open the terminal from PyCharm
• Activate virtualenv :

$> source venv/bin/activate

Terminal command:

Install all python requirements for our project

Create a requirements.txt file in the project and add the python dependencies in it.

numpy
scipy
tensorflow
tensorflow-gpu
keras
ipython
jupyter
six
protobuf==3.1.0
wheel

It should prompt you to install the dependencies, click on Install requirements. Or alternatively you could use the command line to install all the requirements:

pip install -r requirements.txt

 

Packages installed:

Create a Jupyter notebook

Jupyter notebook can be created via PyCharm or directly on the console.

From PyCharm

• Create a Jupyter notebook
• Click on the Play button to connect to Jupyter
• Hit cancel, to create local run configurations
• Click on the notification saying – Run Jupyter notebook
• Optional – open the Jupyter link in browser

 

 

 

 

 

Output:

Directly from console if you only want to use Jupyter:

Hit this command on the console:

jupyter notebook

Then visit the browser with the printer URL on console.

 

Test some sample code

The sample code to test here has been picked from : tensorflow-visualization.

Use the below code on the browser or PyCharm to visualize a graph.

 

 

 

Thats all for this quick post. Hope its helpful.

 

Extra (troubleshooting):

TENSOR Installation Error:

• Dependency on cuda: Refer https://metakermit.com/2017/compiling-tensorflow-with-gpu-support-on-a-macbook-pro/ . Install Cuda dependencies as mentioned in the post.
• Dependency on other python modules: Refer : https://stackoverflow.com/a/42596864/1562474