Migrate the project to AWS about geo-deep-learning

[epeterson12]

To set up the EC2 (Deep Learning AMI (Amazon Linux)) environment that can run our program, run the following commands then edit the configuration file in the EC2 instance. Eventually, the configuration file should be read from an S3 bucket.

source activate pytorch_p36
conda config --add channels conda-forge
conda install torchvision=0.2.1 -y
conda install gdal=2.2.2 -y
conda install ncurses -y
git clone https://github.com/NRCan/geo-deep-learning.git

[epeterson12]

I propose the following process ordering for calling Lambda functions that would run the different stages of the deep learning algorithm. A couple useful references for implementing this workflow are available here https://medium.com/tomincode/launching-ec2-instances-from-lambda-4a96f1264afb and here https://medium.com/swlh/a-guide-to-aws-lambdas-using-python-triggered-by-an-api-call-1e768edd13c5

[epeterson12]

After testing, we have concluded that using Multiple GPUs on AWS isn't worth the extra cost to obtain minimal if any speedup in training (see #17 for further details).