def train(
net: NeuralNet,
train_inputs: np.ndarray,
train_labels: np.ndarray,
input_converter: Callable,
label_converter: Callable,
epoch_count: int = 5000,
batch_size: int = 32,
learning_rate: int = 0.1):
batch_iterator = BatchIterator(train_inputs, train_labels, batch_size)
pbar = tqdm(total=epoch_count)
for epoch in range(epoch_count):
epoch_loss = 0
batch = next(batch_iterator)
for input, label in batch:
vector_input = input_converter(input)
vector_label = label_converter(label)
output = net.predict(vector_input)
epoch_loss += net.loss.loss_func(output, vector_label)
grad = net.loss.grad_func(output, vector_label)
net.backward(grad)
net.gradient_step(learning_rate / batch_size)
pbar.update()
pbar.set_description(desc=f"Training model. Current epoch loss: {round(epoch_loss, 2)}")
