Model Hyperparameter

All hyperparameter values are stored separately from the model code in a YAML file: models/hyperparameters.yaml.
This design avoids hard-coded values and provides a centralized location for managing all hyperparameters.

Format

Hyperparameter values are stored as a list of dictionaries. For example, for m7_ann, the values are stored under the m7 key. Each dictionary contains an hp_no (hyperparameter set ID) and key-value pairs for the parameters.

m7:
    - hp_no: hp1
        seed: 99
        hidden_size: 10
        activation_function: relu
        learning_rate: 0.001
        solver: adam
        epochs: 500
    - hp_no: hp2
        seed: 99
        hidden_size: 10
        activation_function: relu
        learning_rate: 0.01
        solver: adam
        epochs: 500

How to Modify Model Hyperparameter

If, for example, you want to modify the learning rate of the ANN model, create a new hyperparameter set with a new hp_no:

m7:
    - hp_no: hp1
        seed: 99
        hidden_size: 10
        activation_function: relu
        learning_rate: 0.001
        solver: adam
        epochs: 500
    - hp_no: hp2
        seed: 99
        hidden_size: 10
        activation_function: relu
        learning_rate: 0.01
        solver: adam
        epochs: 500
    - hp_no: hp3
        seed: 99
        hidden_size: 10
        activation_function: relu
        learning_rate: 0.1  # modified learning rate
        solver: adam
        epochs: 500

When running experiments, select the desired hyperparameter set by referencing its hp_no, such as hp3, in specs/experiment.yaml under hyperparameter.