Source code for romancal.dq_init.dq_init_step

#! /usr/bin/env python
from __future__ import annotations

from typing import TYPE_CHECKING

import roman_datamodels as rdm
from roman_datamodels.datamodels import RampModel, ScienceRawModel
from roman_datamodels.dqflags import pixel

from romancal.dq_init import dq_initialization
from romancal.stpipe import RomanStep

if TYPE_CHECKING:
    from typing import ClassVar

__all__ = ["DQInitStep"]


[docs] class DQInitStep(RomanStep): """Initialize the Data Quality extension from the mask reference file. The dq_init step initializes the pixeldq attribute of the input datamodel using the MASK reference file. For some Guiding and Image model types, initalize the dq attribute of the input model instead. The dq attribute of the MASK model is bitwise OR'd with the pixeldq (or dq) attribute of the input model. """ class_alias = "dq_init" reference_file_types: ClassVar = ["mask"]
[docs] def process(self, input): """Perform the dq_init calibration step Parameters ---------- input : Roman datamodel input roman datamodel Returns ------- output_model : Roman datamodel result roman datamodel """ # Open datamodel input_model = rdm.open(input) try: input_model = ScienceRawModel.from_tvac_raw(input_model) except ValueError: pass # Convert to RampModel output_model = RampModel.from_science_raw(input_model) # guide window range information x_start = int(input_model.meta.guide_star.window_xstart) x_stop = int(input_model.meta.guide_star.window_xstop) y_start = int(input_model.meta.guide_star.window_ystart) y_stop = int(input_model.meta.guide_star.window_ystop) # set pixeldq array to GW_AFFECTED_DATA (2**4) for the given range output_model.pixeldq[:, x_start:x_stop] = pixel.GW_AFFECTED_DATA self.log.info( f"Flagging rows from: {x_start} to {x_stop} as affected by guide window read" ) output_model.pixeldq[y_start:y_stop, x_start:x_stop] |= pixel.DO_NOT_USE # Get reference file path reference_file_name = self.get_reference_file(output_model, "mask") # Test for reference file if reference_file_name != "N/A" and reference_file_name is not None: # If there are mask files, perform dq step # Open the relevant reference files as datamodels reference_file_model = rdm.open(reference_file_name) self.log.debug(f"Using MASK ref file: {reference_file_name}") # Apply the DQ step, in place dq_initialization.do_dqinit( output_model, reference_file_model, ) # copy original border reference file arrays (data and dq) # to their own attributes. they will also remain attached to # the science data until they are trimmed at ramp_fit # these arrays include the overlap regions in the corners output_model.border_ref_pix_right = output_model.data[:, :, -4:].copy() output_model.border_ref_pix_left = output_model.data[:, :, :4].copy() output_model.border_ref_pix_top = output_model.data[:, :4, :].copy() output_model.border_ref_pix_bottom = output_model.data[:, -4:, :].copy() output_model.dq_border_ref_pix_right = output_model.pixeldq[:, -4:].copy() output_model.dq_border_ref_pix_left = output_model.pixeldq[:, :4].copy() output_model.dq_border_ref_pix_top = output_model.pixeldq[:4, :].copy() output_model.dq_border_ref_pix_bottom = output_model.pixeldq[-4:, :].copy() else: # Skip DQ step if no mask files reference_file_model = None self.log.warning("No MASK reference file found.") self.log.warning("DQ initialization step will be skipped.") output_model.meta.cal_step.dq_init = "SKIPPED" # Close the input and reference files input_model.close() try: reference_file_model.close() except AttributeError: pass if self.save_results: try: self.suffix = "dqinit" except AttributeError: self["suffix"] = "dqinit" return output_model