linear_regression.ipynb Example based on https://www.learnpytorch.io/01_pytorch_workflow/ uses nn.Module as base class and defines own weights

linear_regression_1.ipynb Example based on https://www.learnpytorch.io/01_pytorch_workflow/

• compared to linear_regression.ipynb this uses the nn.linear() class for the linear regression and looks at moving tensors between devices, including creating device agnostice code - i.e. to take a gpu if present but run on cpu if not.

linear_regression_data_loader.ipynb Linear regression example with torch's DataLoader taken from: https://machinelearningmastery.com/using-optimizers-from-pytorch/ with nothing currently added by me.

loading_regression_model2_parameters.ipynb example of loading parameters stored in the models subdirectory

models subdirectory containing saved model parameters.

colab_binary_classification_default_initialisations.ipynb loosely following https://www.learnpytorch.io/02_pytorch_classification/ with help from: https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html#sphx-glr-beginner-blitz-neural-networks-tutorial-py and Andrej Karpathy's youtube channel. Use pytorch's default initialisations for an nn.Linear layer. See also: colab_binary_classification_alternative_initialisations.ipynb

colab_binary_classification_alternative_initialisations.ipynb loosely following https://www.learnpytorch.io/02_pytorch_classification/ with help from: https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html#sphx-glr-beginner-blitz-neural-networks-tutorial-py and Andrej Karpathy's youtube channel. This version uses Kaiming He /Xavier type scaling of the normal distribution for initialisation and zero bias. Compared to the two identical instances of the model with default initialisations, this initialisation converges the model to 100% accuracy on train and validation with 2/3 to a half the number of epochs, and the model trained for the same number of steps has a much smoother decision boundary. See also: colab_binary_classification_default_initialisations.ipynb

colab_multiclass_classification_mini_batch_better_init.ipynb Multiclass classification loosely following https://www.learnpytorch.io/02_pytorch_classification/ with help from: https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html#sphx-glr-beginner-blitz-neural-networks-tutorial-py and Andrej Karpathy's youtube channel. Has an example with full batch and one with mini batches. compared to the binary training examples that I implemented this has better initialization of the output layer, to give more even probability predictions on initialization.

colab_PyG_GCNConv_tutorial_embedding_classification Simple embedding and classification task using GCNConv graphical conv networks on Zachary's Karate club data. Taken from the tutorial at https://colab.research.google.com/drive/1h3-vJGRVloF5zStxL5I0rSy4ZUPNsjy8?usp=sharing#scrollTo=1-W5Kfhu5I-P