net.train() in validation and test phase.

Hi, I'm new to BCNN.
I   don't know why net.train() was used rather than net.eval() in uncertainty_estimation.py when I want to use the model for inference.

> net = getModel(net_type, 1, 10, priors=None, layer_type=layer_type, activation_type=activation_type)
    net.load_state_dict(torch.load(weight_path))
    net.train()
    net.to(device)

Could someone help me about this?