Machine Learning Frameworks Research

Exploration of Machine Learning concepts and popular Python frameworks to explore supervised learnings, basic models, and neural networks.

Description

Module: Python and TensorFlow - Tony Manschula, Henry Shires, and Alex Bashara

(CPR E 487 at Iowa State University): Our deliverables will consist of a modified Lab 1 Jupyter notebook using PyTorch. Additionally, we will also benchmark against TensorFlow in several key areas in a writeup presentation for the final demo:

• Usability
  • Ease of use of the API
  • Clarity and usefulness of API documentation
  • Any lab 1 activities that could not be implemented in PyTorch
• Training accuracy
  • Train a given model using a different number of epochs and report the number that achieved the highest validation accuracy
• Training time/performance
  • What hardware acceleration options do the frameworks support?
  • How much training time did each framework require to achieve its best validation accuracy?
• Resource utilization
  • Memory utilization and how each framework may lend itself to a given selection of hardware (embedded, etc.)

Read entire project report: 487-report.pdf

Module: Scikit-Learn - Henry Shires

My playground using the Scikit-Learn machine learning library for Python, following Google's Machine Learning Recipes by Josh Gordon. I desired to develop a fundamental understanding of machine learning and how it implements into source code to solve problems.

Installation

Create Python Virtual Environment

Guide: https://pytorch.org/get-started/locally/

1. python -m venv venv
2. .\venv\Scripts\activate for Windows or source venv/bin/activate for Unix
3. python -m pip install --upgrade pip
4. pip install -r ./pytorch/requirements.txt or ./tensorflow/requirements.txt or ./scikit-learn/requirements.txt

VS Code Extensions

• Python
• Jupyter
• PyTorch Snippets

Resources

PyTorch Tutorial

Follow instructions from here: https://pytorch.org/tutorials/beginner/basics/intro.html

Use the ./pytorch/tutorial directory of notebooks to get started:

1. Intro
2. Quickstart
3. Tensors
4. Datasets & DataLoaders
5. Transforms
6. Build Model
7. Autograd
8. Optimization
9. Save & Load Model

Scikit-Learn Recipes

• Scikit-Learn Documentation
• Google Developers ML Recipes
• Neural Networks Playground
• Unity 3D ML Agents

PyTorch Equivalent of TF Profiler and Tensorboard

Using the PyTorch Profiler This generates a trace.json file that can be imported into various viewers, the most prominent being Perfetto UI.

Addtitionally, we can directly print inference times in the program using a built-in PyTorch method.