ucsd-cse-253-hw-3

HW3 for CSE 253 (Winter 2018) at UCSD

Getting to and activating computing instance

ssh your_id@ieng6.ucsd.edu

Put in your password.

To activate the computing environment (i don't get what this does)

cs253w

Then depending on whether or not you want to use a GPU.

launch-pytorch.sh

or

launch-pytorch-gpu.sh

You should now have a Docker container where all of your stuff is accessible.

Git

You can clone this repo into either the container started from: launch-pytorch-gpu.sh Or just more generally into your position on the ieng6 cluster.

Running scripts

Problem 1

python network_1.py

python network_2.py

python Network_3.py

These files will produce 2 figures each: percent accuracy and class accuracy.

Problem 2

'TransferLearningFinal.py'

Performs transfer learning; will produce two plots of accuracy and loss, as well as 4 data files for accuracy and loss of training and testing data

'FeatureExtractionFinal.py'

Performs feature extraction after 3rd and 4th layers; will produce 4 plots of accuracy and loss for each of the 3rd and 4th layers, as well as 8 data files for accuracy and loss of training and testing data

'PlotActivations.py'

Plots 5 image inputs as original images, after 1st conv layer, and after last conv layer. Also plots 1st layer weights. Saves all images to a folder (must change directory to run)

Credit

Credit to this repo for the validation data loader.