SlidingWindow¶

class tfsnippet.dataflows.SlidingWindow(data_array, window_size)¶

Bases: tfsnippet.dataflows.data_mappers.DataMapper

DataMapper for producing sliding windows according to indices.

Usage:

data = np.arange(1000)
sw = SlidingWindow(data, window_size=100)

# construct a DataFlow from this SlidingWindow
sw_flow = sw.as_flow(batch_size=64)
# or equivalently
sw_flow = DataFlow.seq(
    0, len(data) - sw.window_size + 1, batch_size=64).map(sw)

Attributes Summary

`data_array`	Get the data array.
`window_size`	Get the window size.

Methods Summary

`__call__`(*arrays)	Transform the input arrays into outputs.
`as_flow`(batch_size[, shuffle, skip_incomplete])	Get a `DataFlow` which iterates through mini-batches of sliding windows upon `data_array`.

Attributes Documentation

data_array¶: Get the data array.

window_size¶: Get the window size.

Methods Documentation

__call__(*arrays)¶

Transform the input arrays into outputs.

Parameters:	*arrays – Arrays to be transformed.
Returns:	The output arrays.
Return type:	tuple[np.ndarray]

as_flow(batch_size, shuffle=False, skip_incomplete=False)¶

Get a DataFlow which iterates through mini-batches of sliding windows upon data_array.

Parameters:	batch_size (int) – Batch size of the data flow. Required. shuffle (bool) – Whether or not to shuffle the numbers before iterating? (default `False`) skip_incomplete (bool) – Whether or not to exclude the last mini-batch if it is incomplete? (default `False`)
Returns:	The data flow for sliding windows.
Return type:	DataFlow