jennyzhang0215/gaan

gaan's Issues

what's the meaning of "incubator-mxnet/src/operator/contrib" in "segs_ops_cuda/README.md"?

can not download the dataset ppi

when i try the command:

python download_data.py --dataset ppi

i get this problem:

urllib.error.URLError: <urlopen error [Errno 111] Connection refused

Failed to recompile mxnet with custom seg_ops

I've followed the README to clone the mxnet and copied GaAN's custom operators. But when I run "make", seg_op.cc returned with serveral errors like:
.../incubator-mxnet/src/operator/contrib/seg_op.cc:371:18: error: ‘NodePtr’ in namespace ‘nnvm’ does not name a type; did you mean ‘NodeEntry’?

.../incubator-mxnet/src/operator/contrib/seg_op.cc:380:2: error: ‘struct mxnet::op::<lambda(int)>’ has no member named ‘add_argument’ 380 | .add_argument("data", "NDArray-or-Symbol", "The input data.")

Any possible solution?
gcc verion: 9.4.0
g++ version 9.4.0
mxnet version: linux_cu112.cmake
cuda version 11.2

brief previus steps:
git clone {mxnet}
cp {seq_op*} {mxnet}/src/contrib
mkdir build
cd build
cmake ..
make -j6

all previous built-in operators had complied fine.

[Question] About USE_SHARPNESS option

Please tell me about the option USE_SHARPNESS. Thank you!

ModuleNotFoundError: No module named 'mxgraph._graph_sampler'

Traceback (most recent call last):
  File "experiments/static_graph/sup_train_sample.py", line 4, in <module>
    from mxgraph.graph import *
  File "/home/augf/GaAN/mxgraph/graph.py", line 4, in <module>
    import mxgraph._graph_sampler as _graph_sampler
ModuleNotFoundError: No module named 'mxgraph._graph_sampler'

AttributeError: module 'mxnet.symbol.contrib' has no attribute 'seg_pool'?

I download the "incubator-mxnet" in the same dir with "GaAN", and copy files as the "README.md". But when i run the "run.sh" in the dir of "cfg_ppi_sup", but it still not works. The problem is:

  File "/home/augf/GaAN/mxgraph/layers/aggregators.py", line 518, in hybrid_forward
    max_pool_data = F.contrib.seg_pool(data=max_pool_data,
AttributeError: module 'mxnet.symbol.contrib' has no attribute 'seg_pool'

And the "README.md" in GraphSampler, what's the meaning of "end points" , "ind_ptr" and "node_ids" and what's the advantage of using this format?
In fact, I want to test the accuary of 'cora' dataset in your model in cpu. Could you give me a python scripts for preprocessing the original dataset and runing the model?