How to add constrain to nn.Parameter?

nlp
1

Currently I implement joint loss model and I can’t find out how to add constrain to nn.Parameters.

self.text_scoreweight = nn.Parameter(torch.FloatTensor([1]))
self.graph_scoreweight = nn.Parameter(torch.FloatTensor([1]))
...

score = self.text_scoreweight * textscore + self.graph_scoreweight * graphscore

I’d like to add constrains of
0 < self.text_scoreweight, self.graph_scoreweight < 1
self.text_scoreweight + self.graph_scoreweight = 1

Could you know how to solve it?

P.S.
I’ll now try this but if there would be a better approach and you could give me an advice, I’d appreciate it.

param1 = nn.Parameter(torch.FloatTensor([1]))
param2 = 1 - param1
Parameterlist.append(param1) #(for L2regularize parameter)
2

Did you manage to solve this problem?
I have also been trying to figure out how to constrain a nn.Parameter for a different multitask problem.