lil-lab/atis

atis

This is the code associated with the paper:

Learning to map context dependent sentences to executable formal queries. Alane Suhr, Srinivasan Iyer, and Yoav Artzi. In NAACL, 2018. paper bib

I will be slowly adding and documenting the code. Please open an issue if you have any questions about the code or getting it to run.

If you want access to the data, code for preprocessing it, and the database to execute the queries, please email me ([email protected]). ATIS comprises LDC93S5, LDC94S19, and LDC95S26.

The main file is run.py. You can train with my parameters by calling train.sh, and evaluate by calling eval.sh (but will need to provide the save file name, and you can edit this script so that it selects segments from gold queries instead, as well as evaluating on different splits of the data).

Memory

Depending on your machine setup, you might not be able to run evaluation on all of the interactions on the GPU. The file cpu_full_interactions.txt contains the list of interactions that will only fit on CPU memory for an NVIDIA Tesla V100 GPU (16GB of RAM per GPU). You can add to this list if you are running into memory problems, and these interactions will be ignored during training and evalution. To predict on them, run evaluation with the --no_gpus flag set to true, and only these interactions will be evaluated (you can concatenate the output files for both runs to evaluate predictions on both sets of interactions).

Prerequisites

You need Crayon installed and running. See this page for details. If you don't want to use Crayon, you can comment it out. You can use the provided scripts remove_experiments.py and delete_all_experiments.py in order to remove old experiments.

atis's Issues

Cannot convert _dynet.Parameters to _dynet.Expression

Hi,

I am running into an issue about dynet:

File "/data/lily/rz268/dialog2sql/atis/interaction_model.py", line 170, in
lambda token: dy.concatenate([self.input_embedder(token), discourse_state]), dropout_amount=self.dropout)
File "_dynet.pyx", line 4366, in _dynet.concatenate
File "_dynet.pyx", line 4382, in _dynet.concatenate
TypeError: Cannot convert _dynet.Parameters to _dynet.Expression

Do you happen to know the reason? Is this related to the DyNet version?

Thanks a lot!

Question about initial SQL generate process

Hi Alane, thanks for your impressive work! I have a question on the section 4.5 of the paper and here is what the paper describes:

The utterance is encoded using the initial discourse state h^I_0 , the discourse state h^I_1 is computed, the query y_1 is generated, and the set of segments S_1 is created.

I want to know the initial query y_1 is generated by the model, or just passed by annotation? If the answer is yes, is it illustrated that the model could translate plain text into SQL alone ?

Thanks for your reply :)

Introduction in configuring and running to application

Hi
We are intreasted in using this application as part of our development. Wanted to know if you could explain how to setup the application and configure our database to it.. and by the way our database is MSSQL database.

FileNotFoundError: [Errno 2] No such file or directory: '../atis_data/data/resplit/processed/train_with_tables.pkl'

I am getting file not found error for "train_with_tables.pkl". Can you please guide me where can I find this pickle file?