PyTorch is a powerful deep learning research platform that provides maximum flexibility and speed, and it allows you to replace numpy to use the power of GPUs. It is similar in some ways to Tensorflow but utilizes dynamic graphs.
To whet your appetite and to get you going in PyTorch, you will learn how to build a simple Convolutional Neural Network on the Fashion-MNIST dataset in this tutorial.
This tutorial assumes that you:
- Installed Anaconda with Python 3.5+
- Updated conda with:
conda update conda - Have a basic understanding of Convolutional Neural Networks (watch this video)
- Create conda environment:
conda create --name pytorch numpy - Activate conda environment:
source activate pytorch - Install PyTorch:
conda install pytorch torchvision -c soumith - Install Jupyter notebook kernel and matplotlib:
conda install nb_conda matplotlib - Deactivate conda environment:
source deactivate
- Either clone this repo or go here and find the section entitled Get the Data located towards the middle.
- (Optional) If you chose to download from the website, download the training and test set images and labels. (note where you saved these files)