DynamicNNBuilder

The DynamicNNbuilder is meant to easily compare different metrics for several different networks. By the use of factories, scaling up the project with more layer type is possible.

The user can provide a directory of folder, to compare in the generated report.

In this README there is a set up guide for how to set up a correct python environment. Furthermore there is a guide to show the usecase of the project.

Project Setup Guide

Prerequisites

• Ensure you have >=Python 3.9.12 installed on your system.

Creating and Activating the Virtual Environment

1. Navigate to Your Project Directory:

   Open a terminal and change to your project's directory.

   cd path/to/your/project

2. Set up the venv

   python3.9 -m venv venv

3. Activate the environment

   venv\Scripts\activate

4. Install the requirementes

   pip install -r requirements.txt

To run the default config file in configs folder use:

python main.py

Arguments

The DynamicNNBuilder uses an ArgumentParser to manage command-line options. Here's a breakdown of all available options:

Option	Short	Type	Default	Description
--file	-f	string	_config_path	Path to the main configuration file (TOML). This can be a single .toml file or a directory containing multiple .toml files. If a directory is provided, the builder will process all .toml files within it.
--save	-s	string	"parameters"	Filename (without extension) to save the trained model's parameters. The saved file will be placed in a directory (likely a "parameters" folder, but this might depend on other configurations).
--load	-l	string	None	Filename (without extension) to load pre-trained model parameters. The code expects the file to be in the same location where parameters are saved. If this option is provided, the model will load these parameters instead of training from scratch.
--kaggle	-k	bool	False	If set to True, the network will be used to generate a CSV prediction file suitable for Kaggle competitions. This assumes the code is set up to handle the Kaggle data format and prediction generation.
--conf_dir	-cd	string	None	Path to a directory containing configuration files (TOML). This is an alternative way to provide configurations, especially useful when comparing multiple network configurations. If this is provided it will overwrite the -f flag.
--report	-r	bool	False	If set to True, a report summarizing the training results and other relevant metrics will be generated.
--report_dir	-rd	string	"report"	Path to the directory where the generated report will be saved.

Example Usage

• Running with a specific TOML file:

  python main.py -f path/to/my_config.toml

!!! UNDER DEVELOPMENT !!!

To easily(well not yet) create and then run toml file, use:

python check.py

This uses the tkinter library to make the toml file you want to run.