resco
Description
resco is a simple python package that report the effect of deep residual learning on backpropagation for 2 neural network architectures:
- A plain linear neural network
- A residual linear neural network
Installation
Install and create a virtualenv
pip install virtualenv
virtualenv .venv
Install the package
pip install .
Execution
resco_analysis --model_name plain_dnn --n_blocks 50 --lr 0.001 --batch_size 256 --n_epochs 5
Gradients of the first neural network layer are reported on tensorboard, you can run your own tensorboard server by using this command
docker image build --tag tensorboard_resco .
docker container run --restart always -d -p 5001:5001 tensorboard_lm