Training a network

Function Signature

def train_model(
    train_inputs: np.ndarray,
    train_targets: np.ndarray,
    test_inputs: np.ndarray,
    test_targets: np.ndarray,
    batch_size: int = 36,
    epochs: int = 500,
    learning_rate: float = 0.1,
    momentum: float = 0.6,
    early_stop: bool = False,
    verbose: bool = True
) -> None:

Parameters

• train_inputs (np.ndarray): The input data to use for training. This should be a numpy array of shape (num_samples, input_shape).

• train_targets (np.ndarray): The target data to use for training. This should be a numpy array of shape (num_samples, output_shape).

• test_inputs (np.ndarray): The input data to use for testing. This should be a numpy array of shape (num_samples, input_shape).

• test_targets (np.ndarray): The target data to use for testing. This should be a numpy array of shape (num_samples, output_shape).

• batch_size (int, default=36): The batch size to use during training.

• epochs (int, default=500): The number of epochs to train the model for.

• learning_rate (float, default=0.1): The learning rate to use during training.

• momentum (float, default=0.6): The momentum to use during training.

• early_stop (bool, default=False): Whether to use early stopping during training. If True, the training will stop when the validation loss stops improving.

• verbose (bool, default=True): Whether to print training progress during training.

Return Value

This function does not return anything. It trains the deeprai.models.FeedForward instance on the given data and saves the updated weights.

The train_model function trains the deeprai.models.FeedForward instance on the given training data using the specified hyperparameters. It also evaluates the model on the test data after each epoch and prints the training progress if verbose=True.

The batch_size parameter specifies the batch size to use during training. The epochs parameter specifies the number of epochs to train the model for. The learning_rate and momentum parameters specify the learning rate and momentum to use during training, respectively.

The early_stop parameter specifies whether to use early stopping during training. If early_stop=True, the training will stop when the validation loss stops improving.

Examples

Here's an example of how to use the train_model function:

from deeprai.models import FeedForward

model = FeedForward()
model.add_dense(784)
model.add_dense(128, activation='relu')
model.add_dense(64, activation='relu')
model.add_dense(10, activation='sigmoid')

train_inputs = ...
train_targets = ...
test_inputs = ...
test_targets = ...

model.train_model(
    train_inputs=train_inputs,
    train_targets=train_targets,
    test_inputs=test_inputs,
    test_targets=test_targets,
    batch_size=32,
    epochs=1000,
    learning_rate=0.1,
    momentum=0.6,
    early_stop=True,
    verbose=True
)

This code creates a FeedForward model with an input shape of (784,), adds three dense layers with ReLU and softmax activation functions, sets