Param hotfix

[hellkite500]

Previously, only single model parameter spaces were fully supported by ngen-cal.

This feature makes it possible to calibrate a multi-bmi model with ngen by allowing multiple parameter sets from different models to be combined into a single search space with a unique (ish) index.

Note

Parameters with the exact same name between two models will be exposed as two parameters in the parameter space, but will always share the same value.

[aaraney]

I don't think this works at least for dds in the uniform case b.c. this does not guarantee unique index values. When dds_update selects a random subset of parameters, if two of the parameters share the same name (e.g. 2 bmi modules have calib param with same name) the following code throw.

ngen-cal/python/ngen_cal/src/ngen/cal/search.py

Lines 77 to 94 in 0f5843d

	for n in neighborhood:
	#permute the variables in neighborhood
	#using a random normal sample * sigma, sigma = 0.2*(max-min)
	#print(n, meta.best_params)
	new = calibration_object.df.loc[n, agent.best_params] + calibration_object.df.loc[n, 'sigma']*np.random.normal(0,1)
	lower = calibration_object.df.loc[n, 'min']
	upper = calibration_object.df.loc[n, 'max']
	#print( new )
	#print( lower )
	#print( upper )
	if new < lower:
	new = lower + (lower - new)
	if new > upper:
	new = lower
	elif new > upper:
	new = upper - (new - upper)
	if new < lower:
	new = upper