Skip to content

Latest commit

 

History

History
13 lines (11 loc) · 779 Bytes

README.md

File metadata and controls

13 lines (11 loc) · 779 Bytes

grad_norms.py

Script for the gradient norms experiments. Script will train a model sequentially for 5 tasks, and calculate the gradients on the memory and test set after 1, 2 and 5 tasks. This is stored at ./results/[DATA]/grad_norms _[RANDOM_IDX].npy as a numpy array with dimensions (iters x 6). The first three elements are on the testset, the next three on the memory. Upon running the plot script, all files with random indices will be merged to the file grad_norms_merged.npy. This file contains all stored runs.

Note: uncomment the necessary lines in the plot line to obtain the desired plots.

Options:

  • --iter ITER : How many iterations should be done (same as running script multiple times)
  • --data DATA: Which dataset to use (mnist, cifar, min)