Ikkuna Neural Network Monitor
Project description
Ikkuna
A tool for monitoring neural network training.
Ikkuna provides a framework for adding live training metrics to your PyTorch model with minimal configuration. It is a PubSub framework which allows practitioners to quickly test metrics implemented against a simple API. The following data is provided
- Activations
- Gradients w.r.t weights and biases
- Gradients w.r.t layer outputs
- Weights
- Biases
- Weight updates
- Bias updates
- Metadata such as current step in the training, current labels and current perdictions
Subscribers consume this data and distill it into metrics. Different backends can be used
- Matplotlib
- Tensorboard
Working with this repository
You should create a conda
envorinment from the provided torch.yaml
file and
pip install -r
the provided requirements.txt
file. You will also have to
install numba
for building the documentation until I have the time to figure
out how to optionally turn off parts of a doc build.
You should also run python setup.py develop
which will install the package
with symlinks to this repository. Since all subscribers are setuptools
plugins, they are
not available unless setup.py
is run.
Documentation
The sphinx-generated html documentation is hosted here.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for ikkuna-0.1.0.post2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | d79fe33270dc63e4db2b1c97254326c6b69a032cf31680c2b8403da2f8ca4fb7 |
|
MD5 | 571a73ecae209202536f7ca7a60515f1 |
|
BLAKE2b-256 | e499a7bdaace433b45e7fdfa845d50f9c56d0127285ddeb2a9e524d47df8f354 |