Check to see if this masking xr.where parallelize

[lsetiawan]

This xr.where needs to be investigated to see if it can parallelize with the use of dask.

echopype/echopype/mask/api.py

Line 500 in 3ca028f

da = xr.where(str2ops[operator](lhs, diff), True, False)

[anantmittal]

Command to run the tests and capture the stdout:
python .ci_helpers/run-test.py --local --pytest-args="-vvrP" echopype/mask/api.py

pytest -vvrP echopype/tests/mask/test_mask.py::test_frequency_differencing

[leewujung]

@lsetiawan : the code you pointed to above was actually not the one we were discussing 😬

Below are the 2 potential places where the calibration code (compute_Sv) code can have bottleneck:

• in get_vend_cal_params_power where a broadcasting is probably used to get a data variable into the "right" shape based on indexing:
  

  echopype/echopype/calibrate/cal_params.py

  Lines 299 to 304 in 593a7e1

  	# Get param dataarray into correct shape
  	da_param = (
  	vend[param]
  	.expand_dims(dim={"ping_time": idxmin["ping_time"]}) # expand dims for direct indexing
  	.sortby(idxmin.channel) # sortby in case channel sequence differs in vend and beam
  	)

• in harmonize_env_param_time where there is either a check for all the timestamps or an interpolation:
  

  echopype/echopype/calibrate/env_params.py

  Lines 53 to 66 in 593a7e1

  	# If there's only 1 time1 value,
  	# or if after dropping NaN there's only 1 time1 value
  	if p["time1"].size == 1 or p.dropna(dim="time1").size == 1:
  	return p.dropna(dim="time1").squeeze(dim="time1").drop("time1")
  	# Direct assignment if all timestamps are identical (EK60 data)
  	elif np.all(p["time1"].values == ping_time.values):
  	return p.rename({"time1": "ping_time"})
  	elif ping_time is None:
  	raise ValueError(f"ping_time needs to be provided for interpolating {p.name}")
  	else:
  	return p.dropna(dim="time1").interp(time1=ping_time)

[leewujung]

@anantmittal : Thanks for investigating the xr.where in the masking code though -- since the change is to use dask.where, the solution would probably be similarly useful for the get_vend_cal_params_power case above.

[leewujung]

I'll close this now, since my comment above is replicated in #1200.