noboard

Sometimes you don't need tensorboard, or any progress board at all. You just need to log some data.

The noboard.csv.SummaryWriter mocks a part of the familiar tensorboard.SummaryWriter interface, but logs data to csv files or other simple common formats.

NOTE: This is an early version. A work in progress. Especially tensor -> scalar conversion.

motivation

• Tensorboard saves its data to a hidden away json file. You could parse that, for much the same effect. But the format seems opaque. This library is a simpler answer that promises its data format is stable, and clear.

• noboard matches the tensorboard.SummaryWriter interface, as best it can, to make it easy to change between them.

dependencies

• python >= 3.6
• numpy

install

git clone [email protected]:parenthetical-e/noboard.git
cd noboard
pip install .

usage

from noboard.csv import SummaryWriter
writer = SummaryWriter(log_dir=None) # creates runs/
for n in range(100):
    writer.add_scalar("reward", reward, n)

tests

Use py.test to run the tests for this package.

git clone [email protected]:parenthetical-e/noboard.git
cd noboard
pip install pytest # if not installed
py.test # should run all tests