zero.Stream.epochs¶

Stream.epochs(max_epoch, epoch_size=None, progress_bar_config={})[source]¶

Iterate over data epochs.

A shortcut for what is probably the most popular form of a training loop in Deep Learning (plus a progress bar):

for epoch in stream.epochs(max_epoch, epoch_size):
    for batch in epoch:
        ...

# is equivalent to:

while stream.epoch < max_epoch:
    stream.increment_epoch()
    for batch in stream.data(epoch_size):
        ...

Parameters

max_epoch (Union[int, float]) – defines the number of epochs. The loop keeps running while self.epoch < max_epoch. If float, must be float('inf') or math.inf.
epoch_size (Optional[Union[int, float]]) – the number of data items in one epoch (is forwarded to Stream.data).
progress_bar_config (Optional[Dict[str, Any]]) – if not None (the default value is {}!), a progress bar for iterations will be displayed and the argument will be interpreted as key-word arguments for tqdm. The following key-word arguments will be automatically added if not presented in progress_bar_config: initial, total (if can be inferred from the arguments and/or from Stream.loader). If [ipywidgets](https://ipywidgets.readthedocs.io) is installed and the program is running in JupyterLab (Jupyter Notebook), then the pretty widget is used instead of the text-based one.

Returns

Iterator over iterators over data from Stream.loader.

Raises

AssertionError – if max_epoch is a finite float or nan.

Return type

Iterator[Iterator[Any]]

Examples

stream = Stream(range(3))
for epoch in stream.epochs(2):
    for x in epoch:
        print(x)
    print('-')

stream = Stream(range(3))
for epoch in stream.epochs(3, 2):
    for x in epoch:
        print(x)
    print('-')