Skip to content

Latest commit

 

History

History
60 lines (46 loc) · 2.6 KB

ORDER_OF_OPERATIONS.md

File metadata and controls

60 lines (46 loc) · 2.6 KB
Raw

Overview

We need to capture Order of Operations (OoO) somewhere, starting here.

This used to be reasonably straightforward and uncomplicated until we added Horizontal ROI (cropping) and Interpolation. Those features introduced a lot of new corner-cases that we didn't adequately think-through at introduction, and which the legacy ENLIGHTEN measurement architecture of Reading and ProcessedReading didn't easily accommodate. In particular, we ended up with a lot of (inconsistent) copy-pasted code, re-applying HorizROI and Interpolation in different places (Controller.display, Measurement.save, Measurements.export, etc).

I'm trying to straighten some of that out now. In short, this is my current vision.

wasatch.Reading

Contains everything we read from the spectrometer (spectrum, temperatures, laser state), with some modest driver-level corrections applied (even/odd fix, bin2x2, bad pixel correction), optionally averaged.

wasatch.ProcessedReading

Retains handle to the original wasatch.Reading, plus adds these key attributes:

  • processed -- where we apply a series of transforms to the spectrum
  • raw -- basically a copy of Reading.spectrum
  • dark -- whatever spectrum was used for dark correction, if any
  • reference -- whatever spectrum was used for trans/refl/abs, if any

More recently, we now store a handle to the SpectrometerSettings used to generate the ProcessedReading, as well as persistent copies of both the wavelengths and wavenumbers arrays.

If enlighten.HorizROIFeature is enabled, it will add a .cropped attribute to the ProcessedReading, which is itself another ProcessedReading object with the cropped (shorter) processed, raw, dark, reference, wavelengths and wavenumbers arrays.

If enlighten.InterpolationFeature is enabled, it will add an .interpolated attribute to the ProcessedReading, which again is itself another ProcessedReading object with the interpolated (may be longer or shorter) versions of the (proc/raw/dark/ref/wl/wn) arrays. The assumption is that if .cropped exists, then .interpolated was built from that, else it was built from the original arrays.

Provides get_foo() accessors for all arrays (proc/raw/dark/ref/wl/wn), which follow the following precedence:

  • .interpolated if found, else
  • .cropped if found, else
  • original arrays

In conclusion, the expectation is that Controller.process_reading will apply HorizROIFeature.process (generating .cropped if enabled), then apply InterpolationFeature.process (generating .interpolated if enabled), and that all subsequent processing (including Measurement.save, Measurements.export etc) should use the ProcessedReading getters.