A regular expression is a pattern for matching and extracting the part of a text string that matters. In Python, re.findall returns matching substrings from an input string. section 4.3.3 section 5.2.5
A practical use is cleaning language-model output. If a generated answer places table data between <BeginTable> and <EndTable>, the pattern '<BeginTable>(.*)<EndTable>' extracts that table before rows and cells are split. section 4.3.3
Extracting a SQL query
When a generated answer surrounds SQL with ```sql and ```, the pattern '```sql(.*)```' finds the query. Python's re.DOTALL flag lets the dot match newlines, which may occur in a multiline query. section 5.2.5
import re
def post_process(raw_answer):
table_text = re.findall( #1
'<BeginTable>(.*)<EndTable>',
raw_answer, re.DOTALL)[0]
results = []
for raw_row in table_text.split('\n'): #2
if raw_row: #3
row = raw_row.split('|')
row = [field.strip() for field in row]
row = [field for field in row if field]
results.append(row)
return results
Shows a complete Python post-processing function that extracts marked table text, splits rows and cells, strips whitespace, and returns lists of values. section 4.3.3
Regular expressions also support simple filename searches:
- Search filenames satisfying Joe*.png to find pictures associated with Joe. section 6.3.1
