Skip to content

Latest commit

 

History

History
62 lines (46 loc) · 3.16 KB

RGB_trainer Code Documentation.md

File metadata and controls

62 lines (46 loc) · 3.16 KB
Raw

Code Documentation

Author: metalmerge

This script performs training and evaluation of a deep learning model using PyTorch's ResNet50 architecture for image classification. The script includes data loading, data augmentation, model training, early stopping, evaluation, and visualization of results.

Libraries Used

  • os: The operating system interface, used for file operations.
  • ssl: Used to disable SSL verification for downloading data.
  • time: Used for timing the execution.
  • torch: The PyTorch deep learning framework.
  • numpy: Used for numerical operations.
  • seaborn: Used for creating visualizations.
  • torch.nn: Contains various neural network layers, loss functions, etc.
  • torch.optim: Contains optimization algorithms.
  • matplotlib.pyplot: Used for creating plots.
  • torch.optim.lr_scheduler: Contains learning rate schedulers.
  • PIL: The Python Imaging Library, used for opening and manipulating images.
  • auto_augment: Custom module for image augmentation.
  • torch.utils.data.DataLoader: Used for loading data efficiently.
  • sklearn.metrics: Contains confusion matrix for evaluation.
  • torchvision.datasets: Contains standard datasets for vision tasks.
  • torchvision.transforms: Contains data transformations.

Variables

  • validation_dataset_path: Path to the validation dataset.
  • training_dataset_path: Path to the training dataset.
  • num_retrain: Number of times to retrain the model.
  • testing: Indicates whether this is a test run ("y" or "n").
  • Various hyperparameter values and settings for model training.

Functions and Execution

  1. The script prompts the user for the number of times to retrain the model (num_retrain) and whether it's a test run (testing).

  2. Based on the test run status, the script defines hyperparameter values for number of epochs, learning rate, and batch size.

  3. The script enters a loop for retraining iterations. Within each iteration:

    • Hyperparameters for training are set based on iteration values.
    • Data transformations are defined using transforms.Compose.
    • Training and validation datasets are loaded using datasets.ImageFolder.
    • The base model (ResNet50) is loaded with pre-trained weights from ImageNet.
    • Model training loop is executed, including loss computation, optimization, and early stopping based on validation loss.
    • Various statistics and visualizations are generated for analysis.

  4. The best model and corresponding statistics are printed after all retraining iterations.

  5. If testing is "y", a system notification is displayed. Otherwise, the script puts the computer to sleep after 5 seconds.

Usage

  1. Run the script in a Python environment.
  2. Follow the prompts to provide retraining iterations and test run status.
  3. The script will perform model training, evaluation, and visualization.

Note

  • This script assumes that you have the required datasets in the specified paths.
  • Make sure to have the necessary libraries (os, ssl, time, torch, numpy, seaborn, matplotlib, PIL, auto_augment, sklearn, torchvision) installed in your environment.
  • Adjust the script as needed for your specific dataset and hyperparameters.