can't reproduce results for beamforming example.

[nbecker]

Hi, thanks for very interesting research. I want to learn more about this topic, so I started by trying to reproduce your results.

I can run main.py in beamforming, but I don't think the results are correct. The values printed for CGCNetRate are negative, which I'm guessing is not correct.

Here is the output I see:

Python 3.11.6 (main, Oct  3 2023, 00:00:00) [GCC 13.2.1 20230728 (Red Hat 13.2.1-1)]
Type 'copyright', 'credits' or 'license' for more information
IPython 8.10.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: 
<<<<<<<<<<<<<2000 layouts: 30_links_1000X1000_2_65_length>>>>>>>>>>>>
<<<<<<<<<<<<<100 layouts: 30_links_1000X1000_2_65_length>>>>>>>>>>>>
-0.004783393088300301 51.881610862819926 -0.00021702632947472804 4.137292114173537
-0.22768607278508673 51.98962966628295 0.0026074801234456384 4.0460363306513045
WMMSE time: 26.840757846832275
WMMSE rate: 101.5578477373576
/home/nbecker/.local/lib/python3.11/site-packages/torch_geometric/deprecation.py:22: UserWarning: 'data.DataLoader' is deprecated, use 'loader.DataLoader' instead
  warnings.warn(out)
/home/nbecker/.local/lib/python3.11/site-packages/torch_geometric/warnings.py:17: UserWarning: The usage of `scatter(reduce='max')` can be accelerated via the 'torch-scatter' package, but it was not found
  warnings.warn(message)
CGCNet time: 0.0028231143951416016
Epoch 001, Train Loss: -59.1929, Val Loss: -66.5330
CGCNet time: 0.002583026885986328
Epoch 002, Train Loss: -68.1153, Val Loss: -72.3139
CGCNet time: 0.002653837203979492
Epoch 003, Train Loss: -74.7163, Val Loss: -79.6695
CGCNet time: 0.0029048919677734375
Epoch 004, Train Loss: -80.1375, Val Loss: -83.0691
CGCNet time: 0.0035305023193359375
Epoch 005, Train Loss: -82.4704, Val Loss: -84.7964
CGCNet time: 0.002895832061767578
Epoch 006, Train Loss: -84.0641, Val Loss: -86.1974
CGCNet time: 0.003092527389526367
Epoch 007, Train Loss: -85.4251, Val Loss: -87.3830
CGCNet time: 0.002747774124145508
Epoch 008, Train Loss: -86.5545, Val Loss: -88.3187
CGCNet time: 0.0032072067260742188
Epoch 009, Train Loss: -87.4401, Val Loss: -89.1301
CGCNet time: 0.0028963088989257812
Epoch 010, Train Loss: -88.0962, Val Loss: -89.6914
CGCNet time: 0.003001689910888672
Epoch 011, Train Loss: -88.6209, Val Loss: -90.0747
CGCNet time: 0.002684354782104492
Epoch 012, Train Loss: -88.9962, Val Loss: -90.4675
CGCNet time: 0.0036978721618652344
Epoch 013, Train Loss: -89.3208, Val Loss: -90.7866
CGCNet time: 0.0029876232147216797
Epoch 014, Train Loss: -89.6851, Val Loss: -91.0316
CGCNet time: 0.0036149024963378906
Epoch 015, Train Loss: -89.9862, Val Loss: -91.2690
CGCNet time: 0.0028505325317382812
Epoch 016, Train Loss: -90.2393, Val Loss: -91.6016
CGCNet time: 0.002826690673828125
Epoch 017, Train Loss: -90.4558, Val Loss: -91.8720
CGCNet time: 0.0033402442932128906
Epoch 018, Train Loss: -90.6925, Val Loss: -92.0400
CGCNet time: 0.00302886962890625
Epoch 019, Train Loss: -90.9140, Val Loss: -92.1601
<<<<<<<<<<<<<50 layouts: 30_links_1000X1000_2_65_length>>>>>>>>>>>>
test size (50, 40, 40, 2) 1154
WMMSE time: 17.919129610061646
WMMSE rate: 126.33086495698453
-0.004783393088300301 51.881610862819926 -0.00021702632947472804 4.137292114173537
-0.22768607278508673 51.98962966628295 0.0026074801234456384 4.0460363306513045
CGCNet time: 0.002907276153564453
CGCNet rate: -114.89688110351562
<<<<<<<<<<<<<50 layouts: 30_links_1000X1000_2_65_length>>>>>>>>>>>>
test size (50, 80, 80, 2) 1632
WMMSE time: 35.70175528526306
WMMSE rate: 227.39076862204732
-0.004783393088300301 51.881610862819926 -0.00021702632947472804 4.137292114173537
-0.22768607278508673 51.98962966628295 0.0026074801234456384 4.0460363306513045
CGCNet time: 0.0034520626068115234
CGCNet rate: -208.24270629882812
<<<<<<<<<<<<<50 layouts: 30_links_1000X1000_2_65_length>>>>>>>>>>>>

I'm testing on python-3.11
torch==2.1.1
torch_geometric==2.4.0