⛵ tf-argonaut

Library for creating visual experiment pipelines in tensorflow. Allows to test out different network concepts against standardized datasets.

The Argonauts were a band of heros and adventures that sailed on their ship Argo through the mediterranean and navigated numerous adventures. Like one of them this library is designed to turn TF into your own argo and navigate your experiments.

Argonaut was originally build to allow easy research experimentation of multi-task settings against common datasets (thereby reducing the overhead required for experimentation).

👶 Getting Started

To install the library simply use PyPi:

pip3 install tf-argonaut

Alternatively you can install the library directly through setup.py:

pip3 install .

You can then import the library:

import argonaut as argo

At its core, argonaut allows you to run experiments with a single line of code and a configuration file (see in folder examples/simple_multitask):

argo.run_experiment("Baseline", "experiment.json", name="SimpleExample")

📜 Concepts

Concepts include:

• Experiments
• Pipeline
• Datasets
• Callbacks

Tools

The library also contains multiple tools that allow to inspect data and quickly start training processes.

💾‍ Coding Examples

TODO: more advanced coding examples

Argonaut also comes with various pre-defined models (although you can also easily plug in every keras model, given right input and output structrue). In particular these models include:

TODO

Debugging

TODO: integrate options for TF2 debugging

⚙ Configuration

Experiments allow you to specify most of the hyper-parameters through a configuration json file. See the detailed configuration guide for more details.

License

This library is provided under the Apache License.

Pull Requests to improve code quality and add new functionality are more then welcome!