Welcome to Pugh Torch’s documentation!¶
Pugh Torch¶
Functions, losses, module blocks, and other pytorch code to facilitate quicker, easier, and more reproduceable deep learning experiments.
Package Features¶
Additional methods to TensorBoard summary writer for adding normalized images and semantic segmentation images.
hetero_cross_entropy for cross_entropy loss across heterogeneous datasets
Convenient dataset downloading/unpacking to
~/.pugh_torch/datasets/
.You can override this via the ENV variable
ROOT_DATASET_PATH
.
Functions to convert string CLI/Hydra/yaml parameters into their respective classes. .. code-block:: guess
Activation functions
Optimizers
Learning rate schedulers
Models
Commonly used building blocks for convolutional neural networks.
Activation-specific weight initializers
Helper MixIns like
LoadStateDictMixin
to provide more user/experiment-friendly functions that upstream pytorch wont implement due to maintaining backwards compatability.
Installation¶
Stable Release: pip install pugh_torch
Development Head: pip install git+https://github.com/BrianPugh/pugh_torch.git
Experiments¶
A big part of this repo is a framework to quickly be able to iterate on ideas.
To accomplish this, we provide the following:
A docker container
brianpugh/pugh_torch
that contains many dependencies experimenters would like to use.You can pull the docker image and launch the container via: .. code-block:: guess
docker pull brianpugh/pugh_torch ./docker_run.sh
This will map
~/.pugh_torch
and the local copy of the git repo into the container. You may change this if you like. This will also pass in any available GPUs and set other common docker flags for running/training neural nets.This container runs a VNC server, incase you need to perform some visual actions, like using
matplotlib.pyplot
A unified training driver
experiments/train.py
to run experiments.From the
experiments/
folder, runpython3 train.py template
to begin training the default resnet50 architecture on ImageNet.ImageNet cannot be automatically downloaded (see the error raised). To get training started with an easier-to-obtain dataset, run: .. code-block:: guess
` python3 train.py template dataset=cifar100 model=cifar100 `
A template project
experiments/template
that should get you going. The goal here is to provide maximum flexibility while minimizing “project startup costs”. We leverage the following libraries:Hydra for managing experiment hyperparameters and other configuration. It’s a good idea to make your code configurable via this configuration rather than directly tweaking code to make experiments more trackable and reproduceable.
PyTorch-Lightning for general project organization and training.
pugh_torch
for various tweaks and helpers that make using the above libraries easier for common projects and tasks.
Documentation¶
For full package documentation please visit BrianPugh.github.io/pugh_torch.
**Free software: MIT license**