README.md

CISC3024 Final Project: Classification of SVHN Dataset with VGG

Huang Yanzhen, DC126732

Mai Jiajun, DC127853

1. Final project root directory is Final_Project. Here are some quick accesses:

   • Source .ipynb file

   • Source .ipynb HTMLPreview

   • Experiment Results Summary Table

   • Huang Yanzhen:

     • Project Slides
     • Project Report PDF

   • Mai Jiajun:

     • Project Slides
     • Project Report PDF

2. Experiment sets/subsets are stored in the .pth files, they include hyper-parameter values, training and validation losses over epochs, and the model state. All of such files are listed in the Google Drive due to the large file size. Create a model folder in the root directory of the final project and put them into it.

Experiment Set Structure:

[ # One experiment set: A list of experiment objects.
    { # One experiment object
        "HYPER_PARAM_0": CANDIDATE_VALUE_0_0,        # Value of the candidate of the first hyper-param
        "HYPER_PARAM_1": CANDIDATE_VALUE_1_0,        # Value of the candidate of the second hyper-param
        "train_losses": train_losses_0,              # List of training losses over epochs
        "valid_losses": valid_losses_0,              # List of validation losses over epochs
        "model_state_dict": exp_model_0.state_dict()    # Model state
    },
    {
        "HYPER_PARAM_0": CANDIDATE_VALUE_0_1,
        "HYPER_PARAM_1": CANDIDATE_VALUE_1_1,
        "train_losses": train_losses_1,
        "valid_losses": valid_losses_1,
        "model_state_dict": exp_model_1.state_dict()
    },
    ...
]

3. Dataset .mat files is also in the Google Drive. Create a data folder in the root directory of the final project and put them into it.