Helper scripts that save the current state of the git repository (including the current commit id at HEAD, and any changes to tracked files) to an output directory. These outputs can then be loaded to restore the git repository to the state it was in when saved.
To save the state of a git repository, run the following command from within the git repository:
./save_git_state.sh /path/to/output/git/state
Optionally, you can specify the path to a git repository:
./save_git_state.sh /path/to/output/git/state /path/to/git/repo
See save_git_state.sh for the list of files saved to the
output directory.
Since this is a shell script, you can run it from your favorite language easily. For example, in python:
# train.py
import subprocess
output_dir = '/path/to/experiment/directory'
subprocess.call(['./save_git_state.sh', output_dir])
# ...
# Run your code
# ...output_dir will then contain the state of your code at the time train.py was
run.
Later, to load the state back, you can use load_git_state.sh. This can be used
to reproduce old experiments.
./save_git_state.sh <git_state_output> <git_repo> <output_dir>
This will load the git state saved in <git_state_output> for the repo in
<git_repo> to <output_dir>.
Under the hood, this actually creates a
worktree in the <output_dir>, so that
it is linked to the original git repository.