Function train_epoch

tensorlayer/utils.py:562–609

Training a given non time-series network by the given cost function, training data, batch_size etc. for one epoch. Parameters ---------- network : TensorLayer Model the network to be trained. X : numpy.array The input of training data y : numpy.array

(
    network, X, y, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None, batch_size=100, shuffle=True
)

def train_epoch(
    network, X, y, cost, train_op=tf.optimizers.Adam(learning_rate=0.0001), acc=None, batch_size=100, shuffle=True
):
    """Training a given non time-series network by the given cost function, training data, batch_size etc.
    for one epoch.

    Parameters
    ----------
    network : TensorLayer Model
        the network to be trained.
    X : numpy.array
        The input of training data
    y : numpy.array
        The target of training data
    cost : TensorLayer or TensorFlow loss function
        Metric for loss function, e.g tl.cost.cross_entropy.
    train_op : TensorFlow optimizer
        The optimizer for training e.g. tf.optimizers.Adam().
    acc : TensorFlow/numpy expression or None
        Metric for accuracy or others. If None, would not print the information.
    batch_size : int
        The batch size for training and evaluating.
    shuffle : boolean
        Indicating whether to shuffle the dataset in training.

    Returns
    -------
    loss_ep : Tensor. Average loss of this epoch.
    acc_ep : Tensor or None. Average accuracy(metric) of this epoch. None if acc is not given.
    n_step : int. Number of iterations taken in this epoch.

    """
    network.train()
    loss_ep = 0
    acc_ep = 0
    n_step = 0
    for X_batch, y_batch in tl.iterate.minibatches(X, y, batch_size, shuffle=shuffle):
        _loss, _acc = _train_step(network, X_batch, y_batch, cost=cost, train_op=train_op, acc=acc)

        loss_ep += _loss
        if acc is not None:
            acc_ep += _acc
        n_step += 1

    loss_ep = loss_ep / n_step
    acc_ep = acc_ep / n_step if acc is not None else None

    return loss_ep, acc_ep, n_step


def run_epoch(network, X, y, cost=None, acc=None, batch_size=100, shuffle=False):