comsm0045's Introduction

PyTorch Labsheets

Labsheets for the Applied Deep Learning course.

Overview

Labsheet	Description
0	Introduction to Python and the scientific Python ecosystem
1	Your First Fully Connected Network
2	BC4 and Your First CNN
3	Techniques for Training DNNs
4	Data Augmentation

If you have trouble viewing the labsheets on github, you can try using the NBViewer service provided by ipython.org.

Environments

In these labs we'll be using two computing environments:

Colaboratory (a hosted version of Jupyter notebooks) for exploring PyTorch and dabbling with simple and non-computationally expensive experiments.
Blue Crystal 4 (docs) for GPU accelerated experiments.

If instead you'd like to install Jupyter locally on your laptop, we provide some guidance on a best efforts basis. If you have trouble setting things up then we'd recommend using Colaboratory instead.

Problems

Kindly file an issue with a description of the problem you're facing, your setup, what you are observing and what you expect to happen instead.

comsm0045's People

Contributors

Watchers

comsm0045's Issues

Fix naming for non full training

Models are currently saved in the same format regardless of non full or full, if loads are saved then its difficult to see which is which

maybe save models from full and non full in different folder?

add a `save_model_params` function to the `Trainer` class

to save the model params as a checkpoint to 'reload' into the model and keep training (more)

Add `get_log_dir` like labsheets

Add a function that gets the log dir name for tensorboard so different runs can be queued on bc4 with different parameters and we can see the difference between them