LiveTune 0.0.8

Python package for ML developers and researchers to change certain variables while their code is executing to make the task of training a ML project easier. This package will allow you to tune some parameters while your code is live from outside of the program.

Dynamic Parameter Tuning 🎶

LiveTune is a cutting-edge Python package that empowers Machine Learning developers and researchers to adjust parameters in real-time while their program is running. Seamlessly modify variables, data, or hyperparameters, such as learning rate and regularizer, without interrupting your code execution. Experience the freedom to fine-tune your ML projects on-the-fly.

Installation

Platform Compatibility

Platform	Status
MacOS 🍎	Supported
Linux 🐧	Supported
Windows 🪟	Not Yet Tested

Installation Guide

1. Ensure you have PyPi installed. If not, follow these instructions.
2. Execute the following command in your terminal:

pip install LiveTune

Key Features

LiveTune offers two primary features: Live Variables and Live Triggers. Both are managed using our intuitive tag system.

Upon starting a program that integrates LiveTune, the terminal will display the port number associated with that program instance.

Feature	Data Types	Description	Use Cases
Live Variable	int, char, string, float, boolean	Modifiable variables during runtime.	Hyperparameters, datasets, user info adjustments, etc.
Live Trigger	boolean	A boolean that, when triggered, returns true for a single call.	Remotely initiate functions, loops, or halt code.

Integration Guide

Setting up Live Variables

A live variable is an integer, char, string, float, or boolean that is able to be changed in the middle of the program by using a command in the terminal or in a different program.

First, import liveVar from LiveTune:

from LiveTune import liveVar

To create a Live Variable, create a new variable in the liveVar class.

myVariable = liveVar(INITIAL_VALUE, 'TAG')

Example usage:

ranking = liveVar(100, 'ranking')

Live Variables can be used just like normal variables of its type, so feel free to call the variable as needed.

To avoid resource loss or unnecessary recompiling, liveVar includes a 'changed' method that returns True or False if the Live Variable has been changed since the last time 'changed' was called in a running instance.

Example usage:

if myVariable.changed():
    print("This variable has changed!")

Note: Derived variables from a Live Variable won't auto-update. For instance, if ranking2 = ranking + 5, modifying ranking won't affect ranking2.

Setting up Live Triggers

A live trigger is a boolean that will always return False when called. If a developer "triggers" it, the boolean will return True the next time it is called. It can be used to trigger functions, and subroutines.

First, import liveVar from LiveTune:

from LiveTune import liveTrigger
myTrigger = liveTrigger('TAG')

Example usage:

stopLoop = liveTrigger('stop')
while(not stopLoop()):
    pass

save = liveTrigger('save')

while training:
    if save():
        save_subroutine()
...

Updating a LiveTune Variable

Upon program initiation, LiveTune assigns a port to your instance. This port links tags to client ports temporarily.

[LiveTune] Port number for the LiveTune dictionary: 61629

To update a variable, use the tune command in the terminal or in an external program.

The tune command has a few settings to ensure the correct variable is changed properly.

Setting	Description	Required	Example
--port / -p	The port number given by LiveTune above, separate per instance.	Always	--port 63573
--tag / -t	The developer-set tag string for the variable. Used to identify the variable.	Always	--tag ranking
--value / -v	The new value for the variable-must match the type of the original (e.g. must be an int if originally an int)	Live Variables only	--value 200
--trigger / -tr	LiveTriggers will not "trigger" if this setting isn't included.	Live Triggers only	--trigger

Example terminal usage:

tune --value 10 -t myVariable -p 13451

tune --tag myTrigger --port 13451 -tr

You can also update livetune variables using their update method. example:

myVariable.update(new_value)

Contributing to LiveTune

Those interested in adding features to the package may submit a pull request. Please keep individual improvements to their own pull requests. Please maintain consistent code format when contributing to ensure the package remains easily understandable for future contributors.

1. Fork the project
2. Create your feature branch git checkout -b my-new-change
3. Commit your changes git commit -am 'Added new feature'
4. Push to the branch git push origin my-new-change
5. Run the Unit Tests specified in the section below
6. Submit a pull request

All contributors should run the Unit Tests below to test their changes.

Unit Tests

This repository contains a tests.py file in the tests folder. Clone this repo by using:

git clone https://github.com/soheilzi/LiveTune.git

Navigate to the LiveTune repository.

cd LiveTune

Ensure Python3 is installed on your computer and run the tests:

python3 LiveTune/tests/tests.py

If all tests pass, LiveTune is working correctly. If contributing, please run all tests before submitting a pull request.

Authors

• Soheil Zibakhsh Shabgahi - [e-mail]
• Aiden Tabrizi - [e-mail]

Contact

If you are interested in using the code for commercial use, please contact [e-mail]

Thank you for using LiveTune!

We're still early in testing, so please fill out the feedback form if you find any bugs or have feature requests.

Thank you for using LiveTune! If it's been beneficial to your work, please consider citing our paper in your publications.

  title={LiveTune: Dynamic Parameter Tuning for Training Deep Neural Networks},
  author={Shabgahi, Soheil Zibakhsh and Sheybani, Nojan and Tabrizi, Aiden and Koushanfar, Farinaz},
  journal={arXiv preprint arXiv:2311.17279},
  year={2023}
}

Your support helps us continue our efforts and assists others in discovering LiveTune. We appreciate your acknowledgment.