Accuracy in cubes

[Levy-Yuwei]

Hi,
I've trained a model with your code in cubes dataset and saved checkpoints every 10 epoches. However, when I test on these checkpoints using test_job_1.py , the output accuracy seems to be around 0.9 from epoch 10 to epoch 790. This is quite different from the result in the paper, which is 94.39%.

The only modifications I made are:

1. move the
   # Create the checkpoint subfolder if nonexistent.
   part below
   # If nonexistent, create the subfolder of the log folder associated with
   # the current training job. Otherwise, verify if a saved checkpoint can
   # be found, and in that case assume that training should continued from
   # it.
   because it originally makes self.__found_job_folder to be always True even when you create a new train task.

2. add 'checkpoint_epoch_frequency':10 in TRAINING_PARAMS

The environment I run your code is:
Ubuntu 18.04.1 LTS
NVIDIA 2080TI
torch related packages:
torch 1.4.0
torch-cluster 1.4.5
torch-geometric 1.3.2
torch-scatter 1.4.0
torch-sparse 0.4.3
torch-spline-conv 1.1.1
torchvision 0.5.0

The accuracy for each 10 epoch is as follow:
10 0.7784522003034902
20 0.881638846737481
30 0.8755690440060698
40 0.834597875569044
50 0.8452200303490136
60 0.8603945371775418
70 0.8877086494688923
80 0.8179059180576631
90 0.8983308042488619
100 0.9089529590288316
110 0.8937784522003035
120 0.9165402124430956
130 0.8998482549317147
140 0.8831562974203339
150 0.8725341426403642
160 0.8877086494688923
170 0.8801213960546282
180 0.834597875569044
190 0.8937784522003035
200 0.8968133535660091
210 0.8937784522003035
220 0.8907435508345979
230 0.9013657056145675
240 0.8968133535660091
250 0.8861911987860395
260 0.8968133535660091
270 0.9089529590288316
280 0.8877086494688923
290 0.8998482549317147
300 0.9028831562974203
310 0.8998482549317147
320 0.8998482549317147
330 0.8952959028831563
340 0.8937784522003035
350 0.9059180576631259
360 0.8877086494688923
370 0.9013657056145675
380 0.8968133535660091
390 0.9074355083459787
400 0.8907435508345979
410 0.8922610015174507
420 0.9028831562974203
430 0.8937784522003035
440 0.8998482549317147
450 0.9013657056145675
460 0.8998482549317147
470 0.9059180576631259
480 0.8725341426403642
490 0.8952959028831563
500 0.8892261001517451
510 0.8983308042488619
520 0.8861911987860395
530 0.8998482549317147
540 0.91350531107739
550 0.8968133535660091
560 0.8952959028831563
570 0.9028831562974203
580 0.866464339908953
590 0.8998482549317147
600 0.8983308042488619
610 0.8907435508345979
620 0.8725341426403642
630 0.9028831562974203
640 0.8573596358118362
650 0.8679817905918058
660 0.9044006069802731
670 0.8755690440060698
680 0.9028831562974203
690 0.9059180576631259
700 0.9089529590288316
710 0.9013657056145675
720 0.8482549317147192
730 0.8952959028831563
740 0.8998482549317147
750 0.8755690440060698
760 0.8937784522003035
770 0.9059180576631259
780 0.8892261001517451
790 0.9074355083459787

Maybe I've missed something, could you pls help me with it?

Best.