A tensorflow-independent tensorboard logger
A small logger that lets you write logs readable by Tensorboard but doesn’t require Tensorflow.
You can use the logger as a context manager:
from tensorboard_easy import Logger import numpy as np with Logger('/path/to/logs/folder/') as log: log.log_scalar('my_scalar', 100, step=1) log.log_image('my_images', np.random.rand(3, 20, 20), step=1)
or you can close the logger explicitly:
log = Logger('/some/other/logs') log.log_text('my_text', "Let's throw in some text", 0) log.log_text('my_text', [['Some', 'tensor'], ['with', 'text!']], 1) log.log_histogram('my_histogram', np.random.rand(500), step=0) log.close()
It supports scalars, images, text and histograms.
You can also create functions, that write to a specific tag and automatically increase the step:
with Logger('/path/to/logs/folder/') as log: write_loss = log.make_log_scalar('loss') for i in range(1, 100): write_loss(1 / i)
It can be installed via pip:
pip install tensorboard-easy
The tensorflow or tensorflow-tensorboard packages are not required, however you will need one of them to visualize your logs.