Tathagat Verma

Model architecture does not have H*W*Q size tensors for (a,b) probability

1

As mentioned in the [paper](https://arxiv.org/pdf/1603.08511.pdf), the loss function uses **Z, Z^** which are of shape `H*W*Q`. However the model architecture computes the `(a,b)` probability distribution in a tensor of shape...

MMD computation - square root missing

1

The `compute_mmd` function in [dist_helper.py](https://github.com/lrjconan/GRAN/blob/43cb4433e6f69401c3a4a6e946ea75da6ec35d72/utils/dist_helper.py#L143) should compute _square root_ of `disc(s1, s1) + disc(s2, s2) - 2*disc(s1, s2)` right? [Here](https://stats.stackexchange.com/q/276618) is a reference for the same.

The `compute_mmd` function in [mmd.py](https://github.com/JiaxuanYou/graph-generation/blob/3444b8ad2fd7ecb6ade45086b4c75f8e2e9f29d1/eval/mmd.py#L80) should compute square root of `disc(s1, s1) + disc(s2, s2) - 2*disc(s1, s2)` right? Referring equation 10 of the [GraphRNN paper](https://arxiv.org/pdf/1802.08773.pdf).