This is an adoption of the code used in a side-by-side comparison of PyTorch and Keras by deepsense.ai.
Since PyTorch and Keras have very different levels of abstraction, this example uses skorch instead which still offers the flexibility of PyTorch but offers an abstraction level that is similar to Keras.
The simplest way is to use conda to install the dependencies:
conda env create -f environment.yml -n avp
source activate avp
pip install git+https://github.com/dnouri/skorch
jupyter-notebook skorch-resnet50.ipynb
After that you should be able to navigate to the notebook URL.