search

caio-freitas / GraphDiffusionImitate

Diffusion-based graph generative policies for imitation learning in robotics tasks 🧠🤖
MIT License
1 stars 1 forks source link

2 problems about GraphARM #61

Closed Jary-lrj closed 6 months ago

Jary-lrj commented 10 months ago

Thank you for your paper and work. But I found 2 problems:

  1. GraphARM does not seem to converge. I used the ZINC data set you provided. The batch_size is 1 and the epoch is 1e4. The final Loss does not converge to a certain value. Is this normal?
  2. When I use my own data set to train GraphARM, there will even be an error that the first few epochs can be trained normally, but then the tensor dimension of edge_index and the dimension of x cannot match. I checked your data processing logic and edge_index.shape[1] must be the square of shape[0] of x. But I will get an error such as when x.shape[0] is 26, edge_index.shape[1]=484 instead of 676. I've checked that my data is correct, does this mean there's something wrong with your code? Looking forward to your reply, thank you again!
caio-freitas commented 6 months ago

GraphARM was moved to another repo