ratishsp / data2text-entity-py Goto Github PK

Hello! I'm trying to reproduce the results of this code and I'm getting the following error:

    main()
  File "train.py", line 494, in main
    train_model(model, fields, optim, data_type, model_opt)
  File "train.py", line 257, in train_model
    train_stats = trainer.train(train_iter, epoch, report_func)
  File "/Users/sanjana/destruct/data2text-entity-py/onmt/Trainer.py", line 176, in train
    self._gradient_accumulation(
  File "/Users/sanjana/destruct/data2text-entity-py/onmt/Trainer.py", line 316, in _gradient_accumulation
    batch_stats = self.train_loss.sharded_compute_loss(
  File "/Users/sanjana/destruct/data2text-entity-py/onmt/Loss.py", line 122, in sharded_compute_loss
    for shard in shards(shard_state, shard_size):
  File "/Users/sanjana/destruct/data2text-entity-py/onmt/Loss.py", line 265, in shards
    torch.autograd.backward(inputs, grads, retain_graph=True)
  File "/Users/sanjana/miniconda3/lib/python3.8/site-packages/torch/autograd/__init__.py", line 125, in backward
    Variable._execution_engine.run_backward(
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [1200, 600]], which is output 0 of TBackward, is at version 2; expected version 1 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

Any guidance would be really appreciated. Thank you! @ratishsp

Hi, when I use your code, there is a problem, it's "Not all arguments have the same value:", which is caused by aeq(sourceL, src_len) in GlobalAttention.py. And I find that length of entities_list and src data , which is 602 and 606 respectively. Apparently, this is because of <unk>,<blank>,</s> and <s>. But when I restrict the length of src data to 602, the model could run normally, but the score of RG/CS/CO is too low.

Please help me .

In the DataSet Section of the ReadMe.
I could not find the file "create_dataset.py" in the boxscore-data
mentioned in
"The input dataset for data2text-plan-py can be created by running the script create_dataset.py in scripts folder."

Hi @ratishsp,

Is it possible to upload the code for creating txt files for training/validation/testing for MLB data?
Something similar to this: https://github.com/ratishsp/data2text-entity-py/blob/master/scripts/create_dataset.py

And also, script for the template-based system on MLB dataset.

Thanks in advance.

Hi,

When I run the evalution with
CUDA_VISIBLE_DEVICES=6 ~/torch/install/bin/th extractor.lua -gpuid 1 -datafile roto-ie.h5 -preddata drive_data/transform_gen/roto_cc-beam5_gens.h5 -dict_pfx "roto-ie" -just_eval, I'm faced with the following error.
/root/torch/install/bin/luajit: extractor.lua:574: bad argument #1 to 'copy' (sizes do not match at /root/torch/extra/cutorch/lib/THC/THCTensorCopy.cu:31)
I print the size and find that the size of p (parameters of generated model) is 2234733, and the size of saved_p (parameters from conv1ie-ep6-94-74.t7) is 2141733. How can I solve the problem?

My torch is installed with LuaJIT. The hdf5 is installed with 'luarocks install hdf5'.

Thank you

Yixian