Glenn, Parker
commited on
Commit
•
2a0d5f7
1
Parent(s):
25b2a12
adding readme
Browse files
README.md
ADDED
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
tags:
|
5 |
+
- text2sql
|
6 |
+
datasets:
|
7 |
+
- splash
|
8 |
+
widget:
|
9 |
+
- text: "Give the name, population, and head of state for the country that has the largest area. || select name, population, continent from country order by surfacearea desc limit 1 || | world_1 | city : id, name, countrycode, district, population | sqlite_sequence : name, seq | country : code, name, continent, region, surfacearea, indepyear, population, lifeexpectancy, gnp, gnpold, localname, governmentform, headofstate, capital, code2 | countrylanguage : countrycode, language, isofficial, percentage || swap continent with head of state because it is not required."
|
10 |
+
---
|
11 |
+
## parkervg/destt5-schema-prediction
|
12 |
+
|
13 |
+
Fine-tuned weights for the schema prediction model described in [Correcting Semantic Parses with Natural Language through Dynamic
|
14 |
+
Schema Encoding](https://arxiv.org/pdf/2305.19974.pdf), based on [t5-large](https://huggingface.co/t5-large).
|
15 |
+
|
16 |
+
|
17 |
+
### Training Data
|
18 |
+
|
19 |
+
The model has been fine-tuned on the 7,481 training examples in the [SPLASH interactive semantic parsing dataset](https://github.com/MSR-LIT/Splash).
|
20 |
+
|
21 |
+
|
22 |
+
### Training Objective
|
23 |
+
|
24 |
+
This model was initialized with [t5-large](https://huggingface.co/t5-large) and fine-tuned with the text-to-text generation objective.
|
25 |
+
|
26 |
+
As this model works in the interactive setting, we utilize the standard text2sql features such as `question` and `db_schema`, in addition to `feedback` and `incorrect_parse`.
|
27 |
+
|
28 |
+
```
|
29 |
+
[question] || [incorrect_parse] || [db_id] | [table] : [column] ( [content] , [content] ) , [column] ( ... ) , [...] | [table] : ... | ... || [feedback]
|
30 |
+
```
|
31 |
+
|
32 |
+
The model then attempts to predict those schema items that appear in the final gold SQL query, prefaced by the `db_id`.
|
33 |
+
|
34 |
+
```
|
35 |
+
[db_id] | [table] : [column] ( [content] , [content] ) , [column] ( ... ) , [...] | [table] : ...
|
36 |
+
```
|
37 |
+
|
38 |
+
|
39 |
+
### Performance
|
40 |
+
|
41 |
+
This model achieves 88.98% F1 score in identifying schema items on the SPLASH test set.
|
42 |
+
|
43 |
+
When combined with the [destt5-text2sql model](https://huggingface.co/parkervg/destt5-text2sql), it achieves 53.43% correction accuracy (exact-match).
|
44 |
+
|
45 |
+
|
46 |
+
### References
|
47 |
+
|
48 |
+
1. [Correcting Semantic Parses with Natural Language through Dynamic
|
49 |
+
Schema Encoding](https://arxiv.org/pdf/2305.19974.pdf)
|
50 |
+
|
51 |
+
2. [DestT5 codebase](https://github.com/parkervg/destt5)
|
52 |
+
|
53 |
+
|
54 |
+
### Citation
|
55 |
+
|
56 |
+
```bibtex
|
57 |
+
@inproceedings{glenn2023correcting,
|
58 |
+
author = {Parker Glenn, Parag Pravin Dakle, Preethi Raghavan},
|
59 |
+
title = "Correcting Semantic Parses with Natural Language through Dynamic Schema Encoding",
|
60 |
+
booktitle = "Proceedings of the 5th Workshop on NLP for Conversational AI",
|
61 |
+
publisher = "Association for Computational Linguistics",
|
62 |
+
year = "2023"
|
63 |
+
}
|
64 |
+
```
|