Enabling GPU execution for neural networks

GPUs operate more slowly than CPUs but have the advantage of performing massive numbers of calculations in parallel. Libraries for the GPU execution of neural training algorithms have been developed and are included in the KNIME Deep Learning – Keras Integration. In this section, we will explore how to speed up the network execution, training, and application with GPU acceleration.

First of all, you need a CUDA-enabled GPU. Then, you need a Python environment that uses the GPU version of the Keras package.

Installing Conda

Conda can be installed easily and for free by going to https://docs.conda.io/projects/conda/en/latest/index.html and downloading the appropriate version for your machine.

First, we set up a GPU Python environment via the dedicated KNIME Preferences page. After that, we can use this Python environment in one of two ways:

• We set it up as the default environment and all workflows will use it for...