Optimizing Performance¶

Selecting sensible propagation parameters¶

Calls to the the propagate() method are almost always the most expensive part of any model. Even the most efficient and streamlined models will suffer from poor performance when working with large arrays and many wavelengths. Understanding how each of the propagate() method’s parameters influence accuracy and speed will allow you to identify appropriate values for your specific needs.

`propagate()`¶

Parameter	Usage	Performance considerations
`wave`, `weight`	Arrays representing wavelength and the corresponding weight of each wavelength in the propagation.	Propagation time scales linearly with the number of wavelengths.
`npix`	Shape of output plane.	The output plane size has minimal impact on propagation performance unless it is astronomically large or if `flatten` is `False`. `npix` should be set to ensure all data is adequately captured by the output plane.
`npix_chip`	Shape of propagation plane.	Propagation time scales quadtatically with `npix_chip` but because values are typically small there are unlikely to be any large opportunities for increasing performance by reducing `npix_chip`.
`oversample`	Number of times to oversample the output and propagation planes.	Propagation time scales quadratically with `oversample`. For accuracy, `oversample` should be selected to ensure propagations are Nyquist sampled, but there is typically no benefit in selecting larger values.
`flatten`	If `True`, the cube of wavelength-dependent output planes is flattened into a single 2D array before being returned. If `False`, a cube of output planes is returned.

`radiometry.trim_spectrum()`¶

Parameter	Usage	Performance considerations
`trim_tol`

`radiometry.sample_spectrum()` and `radiometry.trim_spectrum()`¶

Parameter	Usage	Performance considerations
`wave_sampling`

Multiprocessing¶

Using appropriately sized planes¶

Planes should be sized to ensure the smallest spatial features of interest are adequately sampled.

Image simulation¶

Caching¶

Plane attributes¶

A number of plane attributes are accessed with each propagation wavelength. This behavior does does not impact performance unless any of these attributes are computed on the fly or are otherwise expensive to retrieve. To mitigate any potential performance impacts, Lentil’s propagate() method performs a pre-propagation caching step by calling each plane’s cache_propagate() method, temporarily storing a copy of possibly expensive to compute attributes for faster access during the actual numerical propagation. When the propagation calculations are complete, a post-propagation cleanup calls each plane’s clear_cache_propagate() to clear any cached values.

The cached attributes are defined in a list in each Plane’s cache_attrs attribute. This list is user-settable but the only valid values are ‘amplitude’ and ‘phase’. The default behavior is to cache both amplitude and phase attributes.

DFT matrices¶

Profiling your code¶

There are several approaches to finding bottlenecks and inefficiencies. To really understand what is happening, the code needs to be profiled. The Python standard library includes several profilers. cProfile is simple to use and its profile results files can be visualized using snakeviz.