Revalue Technical Assessment

---

Getting Started

This repository contains a machine learning project designed for multispectral image analysis using PyTorch Lightning. The project is structured to provide an easy-to-use command-line interface for training the model, along with a set of utilities for data processing and model configuration.

Prerequisites

• Python 3.8 or higher
• Poetry for dependency management and packaging.

Installation

1. Clone the repository to your local machine.

2. Install dependencies using Poetry. Navigate to the project directory and run:

   poetry install

   This command reads the pyproject.toml and poetry.lock files to install the necessary dependencies in a virtual environment.

Configuration

Before running the training, you may need to adjust the configurations according to your dataset and training requirements. Configuration options are available in config.py. Review and modify them as necessary to fit your project's needs.

Note, you will also need to create a data directory with eurosat data i.e. data/EuroSAT_MS_Samples

Training the Model

To train the model, use the CLI provided in trainer.py. The CLI supports various options for training customization, such as setting the number of epochs, batch size, and more.

Basic Usage

poetry run python trainer.py [OPTIONS]

Replace [OPTIONS] with your desired command-line options to customize the training session. The available options include:

• --dataset_path: Path to the dataset. Default is specified in config.py.
• --num_epochs: Number of epochs for training. Default is specified in config.py.
• --batch_size: Batch size for training. Default is specified in config.py.
• --learning_rate: Learning rate for the optimizer. Default is specified in config.py.
• --num_input_channels: Number of input channels for the model. Default is specified in config.py.

Example Command

To train your model with custom configurations, you can run the following command:

poetry run python trainer.py --dataset_path "/path/to/dataset" --num_epochs 100 --batch_size 32 --learning_rate 0.001 --num_input_channels 3

Viewing TensorBoard Logs

The training session logs are saved in the tb_logs directory. You can visualize the logs using TensorBoard by running the following command:

 tensorboard --logdir=tb_logs --port=8080

Additional Modules

• dataset.py: Defines the data module for handling the dataset.
• model.py: Contains the definition of the machine learning model.
• utils.py: Provides additional utilities for data processing and model training.