Git Logger for Tensorboard with Pytorch

Features
Dependencies
Installation
Usage
- Logging
- Viewing the Logs

Features

Dependencies

Requires:

python + pip
tensorboard
pytorch
OPTIONAL: tensorflow

Note: Full tensorflow installation is required to view logs in tensorboard. This is currently a requirement for all custom plugins. The plugin will still log without tensorflow, and logs can be viewed in the terminal (see below).

Installation

Ensure you have the above dependencies, then run:

pip install git+ssh://git@github.com/meltzerpete/git_logger_for_tensorboard.git#egg=git_logger_for_tensorboard

You will need to restart your terminal to use the git_logger_view command.

Usage

Logging

With pytorch-lightning:

import pytorch_lightning as pl
from pytorch_lightning.loggers import TensorBoardLogger
from git_logger_for_tensorboard.git_logger import GitLightningLogger

tensorboard_logger = TensorBoardLogger(save_dir='lightning_logs',
                                       name='my_experiment')
tb_logger = TensorBoardLogger(save_dir='lightning_logs',
                              name='my_model')
GitLightningLogger(tb_logger).log()  # all git info/patches are logged to the current lightning run dir

trainer = pl.Trainer(*usual_args,
                     logger=tensorboard_logger)
...

With pytorch only:

from torch.utils.tensorboard import SummaryWriter

from git_logger_for_tensorboard.git_logger import GitLogger

summary_writer = SummaryWriter(log_dir='demo6')
GitLogger(summary_writer).log()  # all git info/patches are logged to the log_dir

...

Viewing the Logs

If tensorflow is installed

Simply go to the GIT_LOGGER tab in tensorboard (see image at top of page).

Clicking a repository will take you there in the browser, clicking a commit hash will copy it to clipboard, and clicking a patch will download it as a file.

If tensorflow is not installed

You can view the logs in the terminal by running the following (ensure you have restarted your shell after installation from pip):

$ git_logger_view --logdir lightning_logs                                                                                                                                                                     main ✚ ✱ ◼

Run: lightning_logs/my_model/version_0
        - branch: main
        - remote_url: git@github.com:meltzerpete/git_logger_for_tensorboard.git
        - last_commit: b7f80b13961c5f66b4c7ca088457c3c979651a1d
        - last_upstream: b7f80b13961c5f66b4c7ca088457c3c979651a1d
        - patch: True
        - upstream_patch: False

...

patch and upstream_patch indicate if a patch is saved (because there were local differences against the commits), and you can view these as follows:

$ git_logger_view --patch lightning_logs/my_model/version_0
#              or --upstream_patch

or save them to file with:

$ git_logger_view --patch lightning_logs/my_model/version_0 > my_diff.patch
#              or --upstream_patch

meltzerpete/git_logger_for_tensorboard