Run for help: step 15

[LMY-nlp0701]

I think the front is running smoothly without bug.

But until it runs to the fifteenth step. I have encountered the following problems.

The order given by the author:
CUDA_VISIBLE_DEVICES=0 th ed/ed.lua -root_data_dir $DATA_PATH -ent_vecs_filename $ENTITY_VECS -model 'global' |& tee log_train_ed

I made changes according to the actual situation:
CUDA_VISIBLE_DEVICES=2 th ed/ed.lua -root_data_dir data_path/ -ent_vecs_filename ent_vecs__ep_69.t7 -model 'global' |& tee log_train_ed

In order to verify that all my previous steps were correct, I gave the results of the fourteenth step.

I hope you can give us a reply.
Thank you!

[octavian-ganea]

Can you please comment the following line and run again ?

https://github.com/dalab/deep-ed/blob/master/entities/pretrained_e2v/e2v.lua#L21

[LMY-nlp0701]

To ensure that entity embeddings has no problem.
I run step 14 again, so I delayed some time.

So, I run the step 15 again. But still encountered the above problems.
So according to your request, 21 lines were noted.
[(https://github.com/dalab/deep-ed/blob/master/entities/pretrained_e2v/e2v.lua#L21)]

But the following problems are encountered.

So I comment 28 line and run again.
[(https://github.com/dalab/deep-ed/blob/master/entities/pretrained_e2v/e2v.lua#L28)]

Happily, the program runs smoothly.

It is worth noting that we still need to install gnuplot to draw, which is not mentioned in your configuration file.
sudo apt-get install gnuplot

Step 15 is not run completed, so I can only provide partial results.
But I think there's nothing wrong with running.

So my question is: what is the meaning of the 21 line and the 28 line assert?
If it is directly annotated, will there be any other impact? For example, will the final experimental results be affected?

[octavian-ganea]

Great, thanks for the extensive details of your run. Lines 21 and 28 just make sure that the embedding corresponding to the UNK entity id has a 0 norm. This should be true according to :
https://github.com/dalab/deep-ed/blob/master/entities/learn_e2v/model_a.lua#L31
but I am not sure why it currently breaks. Hopefully, the entities IDs are not wrongly swapped after training the entity embeddings. I do not expect this to happen, so to affect the results in any way, but please let me know if you observe something different.

[LMY-nlp0701]

Hi, octavian-ganea
Because the program has not run yet, I haven't found any problems yet.
I will tell you if I find any problems after the operation. Thank you!

Thank you for your answer all the time!