2D CNN approach for Table2Vec

• Run python training.py with the following flags:

  • --comment=<insert comment> to add comments.
  • --model_type=<type> for the model type. Currently, there's convE inspired model and Computer Vision inspired model.
  • --gpu=<GPU NO> to choose the gpu rank. Skip when running distributed training.
  • --config=./config_no_conv.toml (Optional) To add additional/overriding configuration file.
  • --distributed to run the distributed training.

eg:

• for distributed training

  • python -m torch.distributed.launch --nproc_per_node=4 training.py -m='cv_insp' --comment='testing all data with new random table gen with/wo distr-gpu' --distributed

• for training on single GPU:2

  • python training.py -g=2 -m='cv_insp' --comment='testing all data with new random table gen with/wo distr-gpu'

• Run tensorboard --logdir outputs/ for the tensorboard.

Data and model files

You can find the data and model dumps to replicate our results through this Google drive link.