LookupTable implementation

[ez81]

I was trying to implement the LookupTable. However the output does not match that from the original torch model. I used the nn.Embedding module and copied the weight tensor from the torch LookupTable. What could be wrong here?

Additional condition in lua_recursive_model

elif name == 'LookupTable':
            n1 = Lambda(lambda x: x.view(1,-1) if 1==len(x.size()) else x )
            n2 = nn.Embedding(m.weight.size(0),m.weight.size(1))
            n2.weight.data.copy_(m.weight)
            n = nn.Sequential(n1,n2)
            add_submodule(seq,n)

Additional condition in lua_recursive_source

elif name == 'LookupTable':
            s1 = 'Lambda(lambda x: x.view(1,-1) if 1==len(x.size()) else x )'
            s2 = 'nn.Embedding({},{})'.format(m.weight.size(0),m.weight.size(1))
            s += ['nn.Sequential({},{}),#LookupTable'.format(s1,s2)]

[zeng9t]

Have you solved that? I have the problem of LookupTable when transfer the lua/torch code to pytorch code. @YZHANGFPE