geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel module

Defines TimeKernel.

Note

Kernel may not be the best name for this, which is a highly overloaded term. The name was chosen based on the use of a statistical kernels or convolutional kernels used in image processing [WikiKernelImageProcessing]. It has nothing to do with kernel methods in optimization or null space in linear algebra.

It is really an ideal time distribution that will be used as a template.

References

[WikiKernelImageProcessing]

https://en.wikipedia.org/wiki/Kernel_(image_processing)

class geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel.TimeKernel(*args, **kwargs)

Bases: ndarray

Represents an ideal relative time sampling pattern.

This is just an ndarray with offsets specified in seconds.

Notes

https://numpy.org/doc/stable/user/basics.subclassing.html#extra-gotchas-custom-del-methods-and-ndarray-base

CommandLine

xdoctest -m geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel TimeKernel --show

Example

>>> from geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel import TimeKernel
>>> from geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel import _random_discrete_relative_times
>>> self = TimeKernel.coerce('-1y,-3m,0,3m,1y')
>>> print(f'self={self!r}')
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autosns()
>>> # Pretend we have discrete observations with the following relative
>>> # time differences
>>> import kwutil
>>> time_range = kwutil.timedelta.coerce('4y')
>>> relative_unixtimes = _random_discrete_relative_times(time_range)
>>> self.plot(relative_unixtimes)
>>> kwplot.show_if_requested()

classmethod coerce(data)

classmethod coerce_multiple(data)

Example

>>> from geowatch.tasks.fusion.datamodules.temporal_sampling.time_kernel import TimeKernel
>>> import ubelt as ub
>>> pattern = ('-3y', '-2.5y', '-2y', '-1.5y', '-1y', 0, '1y', '1.5y', '2y', '2.5y', '3y')
>>> multi_kernel = TimeKernel.coerce_multiple(pattern)
>>> print('multi_kernel = {}'.format(ub.urepr(multi_kernel, nl=2)))

make_soft_mask(relative_unixtimes)

plot(relative_unixtimes)