Custom CNN with Keras

Installing with Conda

This project uses Keras with PlaidML to train and uses TensorFlow to test the model. If we install both TensorFlow and PlaidML on the same virtual environment then TensorFlow will take precedence over PlaidML and we wont be able to use an AMD GPU since it will default to CPU. This is why we should create two different virtual environments:

• veers_custom_cnn_env -> will be used for notebooks 1, 2 and 3
• testing_model_env -> will be used for notebook 2 This also means you will have to add both kernels to jupyter notebook kernels. The steps to do all this are mentioned below.

Clone this project and cd into your project directory.

Run this command to automatically create the conda environment called 'veers_custom_cnn_env' and install all packages

conda env create -f environment_1.yml

Activate this virtual environment

conda activate veers_custom_cnn_env

Add this kernel to jupyter notebook kernels

python -m ipykernel install --user --name='veers_custom_cnn_env'

Now this kernel will appear on jupyter notebook. Deactivate this environment.

conda deactivate

Now install the second environment

conda env create -f environment_2.yml

Activate this virtual environment

conda activate testing_model_env

Add this kernel to jupyter notebook kernels

python -m ipykernel install --user --name='testing_model_env'

Deactivate this environment

conda deactivate

If you run this command, you will see that we have created both environments

conda env list

Running the Project

Activate the first environment

conda activate veers_custom_cnn_env

Set PlaidML as backend

plaidml-setup

Enable experimental device support? (y,n)[n]:n

On most devices 1 would be your CPU, 2 would be the integrated GPU on your CPU and 3 would be your dedicated GPU. Choose 3.

Please choose a default device: 3

Save settings to /Users/veersingh/.plaidml? (y,n)[y]:y

Cd into the project directory

Start the jupyter notebook server

jupyter notebook

Now you can run the notebooks 1,2 and 3. Make sure to set the kernel to 'veers_custom_cnn_env'.

To run notebook 4, when you open it, change the kernel to -> testing_model_env