This was done as part of High Performance Machine Learning course.
Test the performance(Measure the time taken) of ResNet-18 with varying parameters:
Parameters:
- Device: CPU or GPU
- Numworkers
- Type of optimizer
- SGD:simple stochastic gradient descent
- Nesterov
- Adam
- Adagrad
- AdaDelta
- BatchNorm:yes/no
Example command to run file with above arguments:
py lab2.py --device cuda --optimizer sgd --num_workers 8 --batchnorm no
Pytorch,Numpy,Matplotlib
It will give you:
- Time per epoch
- Time for total epochs
Time Measuments
- Total time
- Time to load data
- Time to train
CIFAR-10