Skip to content

Commit 2f9d0f4

Browse files
Ruiqi ZhongRuiqi Zhong
Ruiqi Zhong
authored and
Ruiqi Zhong
committed
updated version of classical dataset
1 parent b3a527a commit 2f9d0f4

File tree

7 files changed

+4087
-764
lines changed

7 files changed

+4087
-764
lines changed

‎README.md‎

Lines changed: 23 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -76,7 +76,7 @@ Include `--progress_bar_for_each_datapoint` if you suspect that the execution go
7676

7777
## Evaluation for Other Classical Text-to-SQL Datasets
7878

79-
**NOTE** Thanks to @rizar we identified several issues with the current release. Please use it only for the purpose of preliminary exploration.
79+
*UPDATE:* we fixed the issue mentioned in https://github.com/taoyds/test-suite-sql-eval/issues/1 . We also added additional features to evaluate on a subset and cache the results to speed up evaluation.
8080

8181
The prior work on classical text-to-sql datasets (ATIS, Academic, Advising, Geography, IMDB, Restaurants, Scholar, Yelp) usually reports the exact string match accuracy and execution accuracy over a single database content, which either exaggerates or deflates the real semantic accuracy.
8282

@@ -89,6 +89,7 @@ All the test datapoints are saved in `classical_test.pkl`. Each test datapoint i
8989
- `variables`: the constants that are used in the SQL query. We also include a field called `ancestor_of_occuring_column`, where we find out all the column that contains this value and recursively find its `ancestor column` (if a column refers to a parent column/has a foreign key reference). This field is especially useful if your algorithm originally uses database content to help generate model predictions.
9090
- `testsuite`: a set of database paths on which we will compare denotation on
9191
- `texts`: the associated natural language descriptions, with the constant value extracted.
92+
- `orig_id`: the original data id from jonathan's repo. it is a tulple of two elements (db_id, idx) - referring to the idx^th element of the list encoded by text2sql-data/data/[db_id].json .
9293

9394
You can evaluate your model in whatever configurations you want. For example, you may choose to plug in the values into the text and ask the model itself to figure out which constants the user has given;
9495
or you can relax the modelling assumption and assume the model has oracle access to the ground truth constant value; or you can further relax the assumption of knowing which "ancestor column" contains the constant provided.
@@ -103,12 +104,31 @@ Suppose you have made a model prediction for every datapoint and write it into a
103104
python3 evaluate_classical.py --gold [gold file] --pred [predicted file] --out_file [output file] --num_processes [process number]
104105
105106
arguments:
106-
[gold file] path to gold file: classical_test.pkl
107+
[gold file] path to gold file. The default is classical_test.pkl, and is hence this argument is optional.
107108
[predicted file] the path to the predicted file. See an example evaluation_examples/classical_test_gold.txt
108109
[output file] the output file path. e.g. goldclassicaltest.pkl
109-
[process number] number of processes to use
110+
[process number] number of processes to use. By default, it is set to cpu_count() // 3, and is hence optional.
111+
[subset] which subset to evaluate on. can be one of {atis,advising,academic,imdb,restaurants,geography,scholar,yelp,full}
112+
[disable_cache] whether to directly apply previously computed result and cache the current results. Use this flag to disable caching.
110113
```
111114

115+
Here is an example command that evaluates the gold prediction file:
116+
117+
```
118+
python3 evaluate_classical.py --pred=evaluation_examples/classical_test_gold.txt --out_file=all_eval_results.json
119+
```
120+
121+
You can also choose to evaluate only on a subset of the datapoints, for example
122+
123+
```
124+
python3 evaluate_classical.py --pred=evaluation_examples/academic_gold.txt --subset=academic --out_file=out/out_academic_test.json
125+
```
126+
127+
By default, the evaluation script will save the results of evaluation in cache.pkl, and use it in the future (since these evaluation take a long time to run).
128+
Use the ``disable_cache`` flag otherwise.
129+
130+
The process through which data are transformed can be seen in classical_provenance.ipynb.
131+
112132

113133
## Citation
114134

‎classical_provenance.ipynb‎

Lines changed: 301 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,301 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "code",
5+
"execution_count": 1,
6+
"metadata": {},
7+
"outputs": [],
8+
"source": [
9+
"import json\n",
10+
"import sqlparse\n",
11+
"import pickle as pkl\n",
12+
"dataset_names = ['academic', 'atis', 'advising', 'geography', 'imdb', 'restaurants', 'scholar', 'yelp']\n",
13+
"\n",
14+
"# these datasets are small, so we use the full set. \n",
15+
"new_split_defined = {'restaurants', 'academic', 'imdb', 'yelp'} "
16+
]
17+
},
18+
{
19+
"cell_type": "code",
20+
"execution_count": 2,
21+
"metadata": {},
22+
"outputs": [],
23+
"source": [
24+
"# loading the original datasets from the paper:\n",
25+
"# Improving Text-to-SQL Evaluation Methodology\n",
26+
"\n",
27+
"# a dataset is a list of dictionaries\n",
28+
"# in the original dictionary, each datapoint might consist of several natural language sentences or SQL\n",
29+
"orig_datasets = []\n",
30+
"for dataset_name in dataset_names:\n",
31+
" orig_dataset = json.load(open('text2sql-data/data/%s.json' % dataset_name))\n",
32+
" for idx, d in enumerate(orig_dataset):\n",
33+
" \n",
34+
" d['orig_id'] = (dataset_name, idx)\n",
35+
" \n",
36+
" # fixing annotations here\n",
37+
" \n",
38+
" # change \"company_name\" to producer name, otherwise there is no variable to replace\n",
39+
" if dataset_name == 'imdb' and idx == 27:\n",
40+
" d['sql'][0] = 'SELECT MOVIEalias0.TITLE FROM COMPANY AS COMPANYalias0 , COPYRIGHT AS COPYRIGHTalias0 , MOVIE AS MOVIEalias0 WHERE COMPANYalias0.NAME = \"producer_name0\" AND COPYRIGHTalias0.CID = COMPANYalias0.ID AND MOVIEalias0.MID = COPYRIGHTalias0.MSID AND MOVIEalias0.RELEASE_YEAR > movie_release_year0 ;'\n",
41+
" \n",
42+
" # removing the extra space surrounding the variable actor_name0\n",
43+
" if dataset_name == 'imdb' and idx == 78:\n",
44+
" d['sql'][0] = 'SELECT MAX( DERIVED_TABLEalias0.DERIVED_FIELDalias0 ) FROM ( SELECT COUNT( DISTINCT ( MOVIEalias0.TITLE ) ) AS DERIVED_FIELDalias0 FROM ACTOR AS ACTORalias0 , CAST AS CASTalias0 , MOVIE AS MOVIEalias0 WHERE ACTORalias0.NAME = \"actor_name0\" AND CASTalias0.AID = ACTORalias0.AID AND MOVIEalias0.MID = CASTalias0.MSID GROUP BY MOVIEalias0.RELEASE_YEAR ) AS DERIVED_TABLEalias0 ;'\n",
45+
" \n",
46+
" # there was a scoping error; changed AUTHORalias1 to AUTHORalias0, PUBLICATIONalias1 to PUBLICATIONalias0\n",
47+
" if dataset_name == 'academic' and idx == 182:\n",
48+
" d['sql'][0] = 'SELECT DERIVED_FIELDalias0 FROM ( SELECT AUTHORalias0.NAME AS DERIVED_FIELDalias0 , COUNT( DISTINCT ( PUBLICATIONalias0.TITLE ) ) AS DERIVED_FIELDalias1 FROM AUTHOR AS AUTHORalias0 , CONFERENCE AS CONFERENCEalias0 , PUBLICATION AS PUBLICATIONalias0 , WRITES AS WRITESalias0 WHERE CONFERENCEalias0.NAME = \"conference_name0\" AND PUBLICATIONalias0.CID = CONFERENCEalias0.CID AND WRITESalias0.AID = AUTHORalias0.AID AND WRITESalias0.PID = PUBLICATIONalias0.PID GROUP BY AUTHORalias0.NAME ) AS DERIVED_TABLEalias0 , ( SELECT AUTHORalias1.NAME AS DERIVED_FIELDalias2 , COUNT( DISTINCT ( PUBLICATIONalias1.TITLE ) ) AS DERIVED_FIELDalias3 FROM AUTHOR AS AUTHORalias1 , CONFERENCE AS CONFERENCEalias1 , PUBLICATION AS PUBLICATIONalias1 , WRITES AS WRITESalias1 WHERE CONFERENCEalias1.NAME = \"conference_name1\" AND PUBLICATIONalias1.CID = CONFERENCEalias1.CID AND WRITESalias1.AID = AUTHORalias1.AID AND WRITESalias1.PID = PUBLICATIONalias1.PID GROUP BY AUTHORalias1.NAME ) AS DERIVED_TABLEalias1 WHERE DERIVED_TABLEalias0.DERIVED_FIELDalias1 > DERIVED_TABLEalias1.DERIVED_FIELDalias3 AND DERIVED_TABLEalias1.DERIVED_FIELDalias2 = DERIVED_TABLEalias0.DERIVED_FIELDalias0 ;'\n",
49+
" \n",
50+
" # wrong number of arguments to function COUNT(), change from \",\" to \"||\" for sqlite3 to recognize and execute\n",
51+
" if dataset_name == 'advising' and idx == 107:\n",
52+
" d['sql'][0] = 'SELECT COUNT( DISTINCT COURSEalias1.DEPARTMENT || COURSEalias0.NUMBER ) FROM COURSE AS COURSEalias0 , COURSE AS COURSEalias1 , COURSE_PREREQUISITE AS COURSE_PREREQUISITEalias0 , STUDENT_RECORD AS STUDENT_RECORDalias0 WHERE COURSEalias0.COURSE_ID = COURSE_PREREQUISITEalias0.PRE_COURSE_ID AND COURSEalias1.COURSE_ID = COURSE_PREREQUISITEalias0.COURSE_ID AND COURSEalias1.DEPARTMENT = \"department0\" AND COURSEalias1.NUMBER = number0 AND STUDENT_RECORDalias0.COURSE_ID = COURSEalias0.COURSE_ID AND STUDENT_RECORDalias0.STUDENT_ID = 1 ;'\n",
53+
" \n",
54+
" # there was not example given for level1 and hence replacing variable with values leads to errors\n",
55+
" if dataset_name == 'advising' and idx == 132:\n",
56+
" d['variables'][0]['example'] = '300'\n",
57+
" \n",
58+
" # cannot use count and order without group by; added grouping by actor_id\n",
59+
" if dataset_name == 'imdb' and idx == 79:\n",
60+
" d['sql'][0] = 'SELECT ACTORalias0.NAME FROM ACTOR AS ACTORalias0 , CAST AS CASTalias0 , MOVIE AS MOVIEalias0 WHERE CASTalias0.AID = ACTORalias0.AID AND MOVIEalias0.MID = CASTalias0.MSID GROUP BY ACTORalias0.AID ORDER BY COUNT( DISTINCT ( MOVIEalias0.TITLE ) ) DESC LIMIT 1 ;'\n",
61+
" \n",
62+
" # cannot use count and order without group by; added grouping by actor_id\n",
63+
" if dataset_name == 'imdb' and idx == 80:\n",
64+
" d['sql'][0] = 'SELECT ACTORalias0.NAME FROM ACTOR AS ACTORalias0 , CAST AS CASTalias0 , DIRECTED_BY AS DIRECTED_BYalias0 , DIRECTOR AS DIRECTORalias0 , MOVIE AS MOVIEalias0 WHERE CASTalias0.AID = ACTORalias0.AID AND DIRECTORalias0.DID = DIRECTED_BYalias0.DID AND MOVIEalias0.MID = CASTalias0.MSID AND MOVIEalias0.MID = DIRECTED_BYalias0.MSID GROUP BY ACTORalias0.AID ORDER BY COUNT( DISTINCT ( MOVIEalias0.TITLE ) ) DESC LIMIT 1 ;'\n",
65+
" \n",
66+
" # table has \"u\" in the neighborhood spelling.\n",
67+
" n_before, n_after = 'NEIGHBORHOOD', 'NEIGHBOURHOOD'\n",
68+
" if dataset_name == 'yelp':\n",
69+
" d['sql'][0] = d['sql'][0].replace(n_before, n_after)\n",
70+
" \n",
71+
" if dataset_name == 'yelp' and idx == 42:\n",
72+
" d['sql'][0] = 'SELECT NEIGHBOURHOODalias0.NEIGHBOURHOOD_NAME FROM BUSINESS AS BUSINESSalias0 , NEIGHBOURHOOD AS NEIGHBOURHOODalias0 , REVIEW AS REVIEWalias0 , USER AS USERalias0 WHERE NEIGHBOURHOODalias0.BUSINESS_ID = BUSINESSalias0.BUSINESS_ID AND REVIEWalias0.BUSINESS_ID = BUSINESSalias0.BUSINESS_ID AND USERalias0.NAME = \"user_name0\" AND USERalias0.USER_ID = REVIEWalias0.USER_ID ;'\n",
73+
"\n",
74+
" orig_datasets.extend(orig_dataset)"
75+
]
76+
},
77+
{
78+
"cell_type": "code",
79+
"execution_count": 3,
80+
"metadata": {},
81+
"outputs": [
82+
{
83+
"name": "stdout",
84+
"output_type": "stream",
85+
"text": [
86+
"There are 3509 datapoints in the new testset\n"
87+
]
88+
}
89+
],
90+
"source": [
91+
"# we create the new testset here\n",
92+
"new_testset = []\n",
93+
"for d in orig_datasets:\n",
94+
" orig_id = d['orig_id']\n",
95+
" db_id, idx = orig_id\n",
96+
" \n",
97+
" # we only incorporate the test split if the dataset is large enough\n",
98+
" # otherwise we incorporate the entire dataset\n",
99+
" if d['query-split'] != 'test' and db_id not in new_split_defined:\n",
100+
" continue\n",
101+
" sql = d['sql'][0]\n",
102+
" instance_variables = d['variables']\n",
103+
" instance_name2examples = {d['name']: d['example'] for d in instance_variables}\n",
104+
" \n",
105+
" # we create a new datapoint for each natural language query\n",
106+
" for sentence in d['sentences']:\n",
107+
" new_datapoint = {\n",
108+
" 'text': sentence['text'],\n",
109+
" 'query': sql,\n",
110+
" 'variables': instance_variables,\n",
111+
" 'orig_id': orig_id,\n",
112+
" 'db_id': db_id,\n",
113+
" 'db_path': 'database/{db_id}/{db_id}.sqlite'.format(db_id=db_id)\n",
114+
" }\n",
115+
" new_testset.append(new_datapoint)\n",
116+
"print('There are %d datapoints in the new testset' % len(new_testset))"
117+
]
118+
},
119+
{
120+
"cell_type": "code",
121+
"execution_count": 4,
122+
"metadata": {},
123+
"outputs": [],
124+
"source": [
125+
"import re\n",
126+
"\n",
127+
"# this block implements a function that extract variable names from text and sql\n",
128+
"# later we use it to ensure that every variable is replaced\n",
129+
"\n",
130+
"variable_pattern = re.compile('^[a-z_]+[0-9]+$')\n",
131+
"\n",
132+
"def extract_variable_names(t):\n",
133+
" tokens = t.replace('\"', '').replace('%', '').split(' ')\n",
134+
" var_names = {v for v in tokens if variable_pattern.match(v) and 'alias' not in v}\n",
135+
" return var_names\n",
136+
"\n",
137+
"test = False\n",
138+
"if test:\n",
139+
" sql = 'SELECT BUSINESSalias0.NAME FROM BUSINESS AS BUSINESSalias0 , REVIEW AS REVIEWalias0 WHERE REVIEWalias0.BUSINESS_ID = BUSINESSalias0.BUSINESS_ID AND REVIEWalias0.MONTH = \"review_month0\" GROUP BY BUSINESSalias0.NAME ORDER BY COUNT( DISTINCT ( REVIEWalias0.TEXT ) ) DESC LIMIT 1 ;'\n",
140+
" print(extract_variable_names(sql))\n",
141+
" text = 'return me the homepage of journal_name0 .'\n",
142+
" print(extract_variable_names(text))"
143+
]
144+
},
145+
{
146+
"cell_type": "code",
147+
"execution_count": 5,
148+
"metadata": {},
149+
"outputs": [],
150+
"source": [
151+
"# this block removes extra space surrounding variable names\n",
152+
"def remove_extra_space_around_variable(t):\n",
153+
" var_names = extract_variable_names(t)\n",
154+
" result = str(t)\n",
155+
" for v in var_names:\n",
156+
" result = result.replace('\" ' + v + ' \"', v)\n",
157+
" return result"
158+
]
159+
},
160+
{
161+
"cell_type": "code",
162+
"execution_count": 6,
163+
"metadata": {},
164+
"outputs": [
165+
{
166+
"name": "stdout",
167+
"output_type": "stream",
168+
"text": [
169+
"set()\n"
170+
]
171+
}
172+
],
173+
"source": [
174+
"problematic = set()\n",
175+
"\n",
176+
"for datapoint in new_testset:\n",
177+
" orig_id = datapoint['orig_id']\n",
178+
" \n",
179+
" # remove extra whitespace surrounding the text\n",
180+
" datapoint['text'] = remove_extra_space_around_variable(datapoint['text'])\n",
181+
" \n",
182+
" # there should not be extra whitespace surrounding the sql variables\n",
183+
" if datapoint['query'] != remove_extra_space_around_variable(datapoint['query']):\n",
184+
" problematic.add(orig_id)\n",
185+
"\n",
186+
" text_vars = extract_variable_names(datapoint['text'])\n",
187+
" sql_vars = extract_variable_names(datapoint['query'])\n",
188+
" \n",
189+
" instance_variables = {d['name']: d for d in datapoint['variables']}\n",
190+
" \n",
191+
" # we ensure that all the variables in the sql query and the text can be replaced\n",
192+
" # by some variable in the variable dictionary\n",
193+
" if len(text_vars - instance_variables.keys()) != 0 or len(sql_vars - instance_variables.keys()):\n",
194+
" problematic.add(orig_id)\n",
195+
" \n",
196+
" # replace the variables with the examples in the variable dictionary\n",
197+
" for text_var in text_vars:\n",
198+
" datapoint['text'] = datapoint['text'].replace(text_var, instance_variables[text_var]['example'])\n",
199+
" \n",
200+
" for sql_var in sql_vars:\n",
201+
" datapoint['query'] = datapoint['query'].replace(sql_var, instance_variables[sql_var]['example'])\n",
202+
"\n",
203+
"# we can trace back which datapoints do not satisfy the assumption,\n",
204+
"# then go back and fix it manually\n",
205+
"print(problematic)"
206+
]
207+
},
208+
{
209+
"cell_type": "code",
210+
"execution_count": 7,
211+
"metadata": {},
212+
"outputs": [
213+
{
214+
"name": "stdout",
215+
"output_type": "stream",
216+
"text": [
217+
"[{'db_id': 'academic',\n",
218+
" 'db_path': 'database/academic/academic.sqlite',\n",
219+
" 'orig_id': ('academic', 0),\n",
220+
" 'query': 'SELECT JOURNALalias0.HOMEPAGE FROM JOURNAL AS JOURNALalias0 WHERE '\n",
221+
" 'JOURNALalias0.NAME = \"PVLDB\" ;',\n",
222+
" 'text': 'return me the homepage of PVLDB .',\n",
223+
" 'variables': [{'example': 'PVLDB',\n",
224+
" 'location': 'both',\n",
225+
" 'name': 'journal_name0',\n",
226+
" 'type': 'journal_name'}]},\n",
227+
" {'db_id': 'academic',\n",
228+
" 'db_path': 'database/academic/academic.sqlite',\n",
229+
" 'orig_id': ('academic', 1),\n",
230+
" 'query': 'SELECT AUTHORalias0.HOMEPAGE FROM AUTHOR AS AUTHORalias0 WHERE '\n",
231+
" 'AUTHORalias0.NAME = \"H. V. Jagadish\" ;',\n",
232+
" 'text': 'return me the homepage of H. V. Jagadish .',\n",
233+
" 'variables': [{'example': 'H. V. Jagadish',\n",
234+
" 'location': 'both',\n",
235+
" 'name': 'author_name0',\n",
236+
" 'type': 'author_name'}]},\n",
237+
" {'db_id': 'academic',\n",
238+
" 'db_path': 'database/academic/academic.sqlite',\n",
239+
" 'orig_id': ('academic', 2),\n",
240+
" 'query': 'SELECT PUBLICATIONalias0.ABSTRACT FROM PUBLICATION AS '\n",
241+
" 'PUBLICATIONalias0 WHERE PUBLICATIONalias0.TITLE = \"Making database '\n",
242+
" 'systems usable\" ;',\n",
243+
" 'text': 'return me the abstract of Making database systems usable .',\n",
244+
" 'variables': [{'example': 'Making database systems usable',\n",
245+
" 'location': 'both',\n",
246+
" 'name': 'publication_title0',\n",
247+
" 'type': 'publication_title'}]},\n",
248+
" {'db_id': 'academic',\n",
249+
" 'db_path': 'database/academic/academic.sqlite',\n",
250+
" 'orig_id': ('academic', 3),\n",
251+
" 'query': 'SELECT PUBLICATIONalias0.YEAR FROM PUBLICATION AS '\n",
252+
" 'PUBLICATIONalias0 WHERE PUBLICATIONalias0.TITLE = \"Making database '\n",
253+
" 'systems usable\" ;',\n",
254+
" 'text': 'return me the year of Making database systems usable',\n",
255+
" 'variables': [{'example': 'Making database systems usable',\n",
256+
" 'location': 'both',\n",
257+
" 'name': 'publication_title0',\n",
258+
" 'type': 'publication_title'}]},\n",
259+
" {'db_id': 'academic',\n",
260+
" 'db_path': 'database/academic/academic.sqlite',\n",
261+
" 'orig_id': ('academic', 3),\n",
262+
" 'query': 'SELECT PUBLICATIONalias0.YEAR FROM PUBLICATION AS '\n",
263+
" 'PUBLICATIONalias0 WHERE PUBLICATIONalias0.TITLE = \"Making database '\n",
264+
" 'systems usable\" ;',\n",
265+
" 'text': 'return me the year of Making database systems usable .',\n",
266+
" 'variables': [{'example': 'Making database systems usable',\n",
267+
" 'location': 'both',\n",
268+
" 'name': 'publication_title0',\n",
269+
" 'type': 'publication_title'}]}]\n"
270+
]
271+
}
272+
],
273+
"source": [
274+
"from pprint import pprint\n",
275+
"\n",
276+
"pprint(new_testset[:5])"
277+
]
278+
}
279+
],
280+
"metadata": {
281+
"kernelspec": {
282+
"display_name": "Python 3",
283+
"language": "python",
284+
"name": "python3"
285+
},
286+
"language_info": {
287+
"codemirror_mode": {
288+
"name": "ipython",
289+
"version": 3
290+
},
291+
"file_extension": ".py",
292+
"mimetype": "text/x-python",
293+
"name": "python",
294+
"nbconvert_exporter": "python",
295+
"pygments_lexer": "ipython3",
296+
"version": "3.7.4"
297+
}
298+
},
299+
"nbformat": 4,
300+
"nbformat_minor": 2
301+
}

‎classical_test.pkl‎

2.48 MB
Binary file not shown.

‎database/readme.txt‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,2 +1,2 @@
1-
Please download the database from here: https://drive.google.com/file/d/1IJvpd30D3qP6BZu_1bwUSi7JCyynEOMp/view?usp=sharing and decompress in this directory.
2-
After this step, "test-suite-sql-eval/database/atis/atis.sqlite" should be a valid file path.
1+
Please download the database from the goolge drive link mentioned in the repo-level readme and decompress in this directory.
2+
After this step, "test-suite-sql-eval/database/atis/atis.sqlite" should be a valid file path.

0 commit comments

Comments
 (0)