TWIRLSUnfoldingAndAttentionο
- class dgl.nn.pytorch.conv.TWIRLSUnfoldingAndAttention(d, alp, lam, prop_step, attn_aft=-1, tau=0.2, T=-1, p=1, use_eta=False, init_att=False, attn_dropout=0, precond=True)[source]ο
Bases:
ModuleDescriptionο
Combine propagation and attention together.
- param d:
Size of graph feature.
- type d:
int
- param alp:
Step size. \(\alpha\) in ther paper.
- type alp:
float
- param lam:
Coefficient of graph smooth term. \(\lambda\) in ther paper.
- type lam:
int
- param prop_step:
Number of propagation steps
- type prop_step:
int
- param attn_aft:
Where to put attention layer. i.e. number of propagation steps before attention. If set to
-1, then no attention.- type attn_aft:
int
- param tau:
The lower thresholding parameter. Correspond to \(\tau\) in the paper.
- type tau:
float
- param T:
The upper thresholding parameter. Correspond to \(T\) in the paper.
- type T:
float
- param p:
Correspond to \(\rho\) in the paper..
- type p:
float
- param use_eta:
If True, learn a weight vector for each dimension when doing attention.
- type use_eta:
bool
- param init_att:
If
True, add an extra attention layer before propagation.- type init_att:
bool
- param attn_dropout:
the dropout rate of attention value. Default:
0.0.- type attn_dropout:
float
- param precond:
If
True, use pre-conditioned & reparameterized version propagation (eq.28), else use normalized laplacian (eq.30).- type precond:
bool
Example
>>> import dgl >>> from dgl.nn import TWIRLSUnfoldingAndAttention >>> import torch as th
>>> g = dgl.graph(([0, 1, 2, 3, 2, 5], [1, 2, 3, 4, 0, 3])).add_self_loop() >>> feat = th.ones(6,5) >>> prop = TWIRLSUnfoldingAndAttention(10, 1, 1, prop_step=3) >>> res = prop(g,feat) >>> res tensor([[2.5000, 2.5000, 2.5000, 2.5000, 2.5000], [2.5000, 2.5000, 2.5000, 2.5000, 2.5000], [2.5000, 2.5000, 2.5000, 2.5000, 2.5000], [3.7656, 3.7656, 3.7656, 3.7656, 3.7656], [2.5217, 2.5217, 2.5217, 2.5217, 2.5217], [4.0000, 4.0000, 4.0000, 4.0000, 4.0000]])