AI-Weather-Classification

Table of Contents

• Project Description
• Installation
• Usage
  • Training
  • Optimize Hyperparameters
  • Evaluation
  • Prediction

Project Description

This project is an implementation of a neural network using PyTorch to classify the weather conditions in images. It uses machine learning techniques to accurately predict weather conditions based on the input images.

Installation

To install the project, make sure you have installed a conda.

Look at the conda documentation for more information.

Now, follow these steps to install the project:

1. Clone the repository:

git clone https://github.com/Nicolas2912/AI-Weather-Classification.git
cd AI-Weather-Classification

2. Create a new conda environment:

conda create -n weather-classification python=3.9.13

3. Activate the conda environment:

conda activate weather-classification

4. Install the required packages:

pip install -r requirements.txt

Usage

Training

To train you own model run the following command:

python src/training/train_model.py

This will train a model using the images in the dataset folder with the default arguments. The model with name weather-model will be saved in the models folder. Following arguments can be specified:

• --data: Path to the dataset folder. Default: dataset directory within the project root
• --device: Device to train the model on. Default: cuda if available, else cpu. Allowed values: cpu, cuda, mps.
• --epochs: Number of epochs to train the model. Default: 28.
• --model_path: Path to save the trained model. Default: models/weather-model.pth.
• --verbose: Print model training progress and other model details. Default: True.

After finishing the training, the model will also be evaluated on a test dataset. The evaluation results will be printed to the console. The training results are saved in the experiments folder.

Example usage with arguments:

python src/training/train_model.py --data dataset --device cuda --epochs 30 --model_path models/my-weather-model.pth --verbose True

Optimize Hyperparameters

To optimize the hyperparameters of the model run the following command:

python src/training/optimize_hyperparameters.py

This will optimize the hyperparameters of the model using optuna.

Following arguments can be specified:

• --data: Path to the dataset folder. Default: dataset directory within the project root.
• --n_trials: Number of optimization trials to run. Default: 50.
• --epochs: Number of epochs for training during each trial. Default: 10.
• --device: Device to train the model on. Default: cuda if available, else cpu. Allowed values: cpu, cuda, mps.
• --mode: Mode for optimization, indicating the scale of the optimization process. Default: small. Allowed values: small, large.
• --study_name: Name of the study to save the optimization results. Default: Automatically generated name based on the current date and time.

If mode==large you don't need to specify the epochs since it is a hyperparameter that will be optimized.

You can inspect the optimization process withoptuna-dashboard. Make sure optuna-dashboard is installed. You can install it with:

pip install optuna-dashboard

After the package is installed, you can start the dashboard with:

optuna-dashboard sqlite:///db.sqlite3

Example usage with arguments:

python src/training/optimize_hyperparameters.py --data dataset --n_trials 100 --epochs 20 --device cuda --mode small --study_name my-study

Evaluation

To evaluate the model run the following command:

python src/evaluation/evaluation.py

This will evaluate the model on the test dataset. The evaluation results will be printed to the console and the confusion matrix will be saved in the experiments folder.

Following arguments can be specified:

• --data: Path to the dataset folder. Default: dataset directory within the project root.
• --model: Path to the model file. Default: models/weather-model.pth.
• --device: Device to run the evaluation on. Default: cuda if available, else cpu. Allowed values: cpu, cuda, mps.

Example usage with arguments:

python src/evaluation/evaluation.py --data dataset --model models/my-weather-model.pth --device cuda

Prediction

To predict the weather condition of an image run the following command:

python src/prediction/predict.py --image_path path/to/image.jpg

This will predict the weather condition of the image or images specified by image_path using the trained model. The prediction results will be printed to the console.

Following arguments can be specified:

• --model: Path to the trained model file. Default: models/weather-model.pth.
• --images: Path the directory containing classes of the images. Default: test_image directory within the project root.
• --dataset: Path to the dataset directory used to initialize the dataset structure and classes. Default: dataset directory within the project root.
• --device: Device to run the prediction on. Default: cuda if available, else cpu. Allowed values: cpu, cuda, mps.

Example usage with arguments:

python src/prediction/predict.py --model models/my-weather-model.pth --images test_image --dataset dataset --device cuda