Regarding the significance of constants I1,I2,I3 and wres

[dashp7]

In the file robertson_pinn/robertson_pinn.py, there is a code snippet:

Stiff-PINN/robertson_pinn/robertson_pinn.py

Lines 128 to 132 in ec41e5e

	if use_annealing == False:
	scaling_fac = 1e-7
	I1 = torch.Tensor([1,0,0]).to(device=device) * scaling_fac
	I2 = torch.Tensor([0,1e-9,0]).to(device=device) * scaling_fac
	I3 = torch.Tensor([0,0,1e5]).to(device=device) * scaling_fac

Moving to :

Stiff-PINN/robertson_pinn/robertson_pinn.py

Lines 179 to 191 in ec41e5e

	if use_annealing == False:
	loss_res_train = I1 * criterion(dydt_train[:,0] * w_res[0], rhs_train[:,0] * w_res[0]) \
	+ I2 * criterion(dydt_train[:,1] * w_res[1], \
	rhs_train[:,1] * w_res[1]) + \
	I3 * criterion(dydt_train[:,2] * w_res[2], \
	rhs_train[:,2] * w_res[2])
	loss_res_test = I1 * criterion(dydt_test[:,0] * w_res[0], rhs_test[:,0] * w_res[0]) \
	+ I2 * criterion(dydt_test[:,1] * w_res[1], \
	rhs_test[:,1] * w_res[1]) + \
	I3 * criterion(dydt_test[:,2] * w_res[2], \
	rhs_test[:,2] * w_res[2])

It can be seen that two sets of weights have been used, for example, I1 * criterion(dydt_train[:,0] * w_res[0], rhs_train[:,0] * w_res[0]) for the first species. Can you please explain why two separate sets of weights(I and wres) have been used?