Optional FixedPointSampling without Replacement

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

……
    def forward(self, data):
        pos, batch = data.pos, data.batch
        
        N = len(pos)
        psid, ptid = knn_graph(pos, self.k, batch, loop=False, flow=self.flow) # edge_index:[E, 2]
……

Expected behavior

N * K == len(ptid)

Environment

• OS: Colab Linux
• Python version: 3
• PyTorch version: 1.2.0
• CUDA/cuDNN version: 10.0
• GCC version:
• Any other relevant information:

Additional context

data comes from shapenet and FixedPoints(2048)

[WMF1997]

hello @rusty1s: it is just a type error in the batch line. .long()) should be .long() he typed one more ‘)’

and the code should be

with my result.

and hello @cy69855522

In [1]: import torch_geometric
INFO:rdflib:RDFLib Version: 4.2.2
D:\ProgramData\Anaconda3\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters

In [2]: import torch_geometric.transforms as gtransforms

In [3]: import torch_geometric.datasets as gdatasets

In [4]: m10 = gdatasets.ModelNet('./ModelNet/ModelNet10/')

In [5]: m10_0 = m10[0]

In [6]: m10_0
Out[6]: Data(face=[3, 30734], pos=[18472, 3], y=[1])

In [7]: knn = gtransforms.KNNGraph(k=4)

In [8]: mm = knn(m10_0)

In [9]: mm
Out[9]: Data(edge_index=[2, 73888], face=[3, 30734], pos=[18472, 3], y=[1])

In [10]: 18472*4
Out[10]: 73888

In my code, I do not think there is a question on knn_graph’s flow is source_to_target, I picked up the index, showing that it is okay.

In [15]: mm.edge_index[1][:14]
Out[15]: tensor([0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3])

and I do not test the batch, perhaps it is the same.

yours sincerely, @wmf1997

[rusty1s]

Yes, you are right 😃 Sorry!