11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293 | class MarkdownElementNodeParser(BaseElementNodeParser):
"""
Markdown element node parser.
Splits a markdown document into Text Nodes and Index Nodes corresponding to embedded objects
(e.g. tables).
"""
@classmethod
def class_name(cls) -> str:
return "MarkdownElementNodeParser"
def get_nodes_from_node(self, node: TextNode) -> List[BaseNode]:
"""Get nodes from node."""
elements = self.extract_elements(
node.get_content(), table_filters=[self.filter_table], node_id=node.node_id
)
elements = self.extract_html_tables(elements)
table_elements = self.get_table_elements(elements)
# extract summaries over table elements
self.extract_table_summaries(table_elements)
# convert into nodes
# will return a list of Nodes and Index Nodes
nodes = self.get_nodes_from_elements(
elements, node, ref_doc_text=node.get_content()
)
source_document = node.source_node or node.as_related_node_info()
for n in nodes:
n.relationships[NodeRelationship.SOURCE] = source_document
n.metadata.update(node.metadata)
return nodes
async def aget_nodes_from_node(self, node: TextNode) -> List[BaseNode]:
"""Get nodes from node."""
elements = self.extract_elements(
node.get_content(), table_filters=[self.filter_table], node_id=node.node_id
)
table_elements = self.get_table_elements(elements)
# extract summaries over table elements
await self.aextract_table_summaries(table_elements)
# convert into nodes
# will return a list of Nodes and Index Nodes
nodes = self.get_nodes_from_elements(
elements, node, ref_doc_text=node.get_content()
)
source_document = node.source_node or node.as_related_node_info()
for n in nodes:
n.relationships[NodeRelationship.SOURCE] = source_document
n.metadata.update(node.metadata)
return nodes
def extract_html_tables(self, elements: List[Element]) -> List[Element]:
"""
Extract html tables from text.
Returns:
List[Element]: text elements split by table_text element
"""
new_elements = []
for element in elements:
if element.type != "text":
# skip when it is not text
new_elements.append(element)
continue
else:
text = element.element
last_pos = 0
i = 0
n = len(text)
while i < n:
table_start = text.find("<table>", i)
if table_start == -1:
break
table_end = text.find("</table>", table_start)
if table_end - table_start <= 7:
# not a valid <table></table>
break
# extract text before the table
pre_text = text[last_pos:table_start].strip()
if pre_text:
new_elements.append(
Element(
id=f"{element.id}_{len(new_elements)}",
type="text",
element=pre_text,
)
)
# extract the html table
table_content = text[
table_start : table_end + 8
] # 8 is length of </table>
new_elements.append(
Element(
id=f"{element.id}_{len(new_elements)}",
type="table_text",
element=table_content,
)
)
last_pos = table_end + 8
i = last_pos
# add the last piece of text
final_text = text[last_pos:].strip()
if final_text:
new_elements.append(
Element(
id=f"{element.id}_{len(new_elements)}",
type="text",
element=final_text,
)
)
return new_elements
def extract_elements(
self,
text: str,
node_id: Optional[str] = None,
table_filters: Optional[List[Callable]] = None,
**kwargs: Any,
) -> List[Element]:
# get node id for each node so that we can avoid using the same id for different nodes
"""Extract elements from text."""
lines = text.split("\n")
currentElement = None
elements: List[Element] = []
# Then parse the lines
for line in lines:
if line.startswith("```"):
# check if this is the end of a code block
if currentElement is not None and currentElement.type == "code":
elements.append(currentElement)
currentElement = None
# if there is some text after the ``` create a text element with it
if len(line) > 3:
elements.append(
Element(
id=f"id_{len(elements)}",
type="text",
element=line.lstrip("```"),
)
)
elif line.count("```") == 2 and line[-3] != "`":
# check if inline code block (aka have a second ``` in line but not at the end)
if currentElement is not None:
elements.append(currentElement)
currentElement = Element(
id=f"id_{len(elements)}",
type="code",
element=line.lstrip("```"),
)
elif currentElement is not None and currentElement.type == "text":
currentElement.element += "\n" + line
else:
if currentElement is not None:
elements.append(currentElement)
currentElement = Element(
id=f"id_{len(elements)}", type="text", element=line
)
elif currentElement is not None and currentElement.type == "code":
currentElement.element += "\n" + line
elif line.startswith("|"):
if currentElement is not None and currentElement.type != "table":
if currentElement is not None:
elements.append(currentElement)
currentElement = Element(
id=f"id_{len(elements)}", type="table", element=line
)
elif currentElement is not None:
currentElement.element += "\n" + line
else:
currentElement = Element(
id=f"id_{len(elements)}", type="table", element=line
)
elif line.startswith("#"):
if currentElement is not None:
elements.append(currentElement)
currentElement = Element(
id=f"id_{len(elements)}",
type="title",
element=line.lstrip("#"),
title_level=len(line) - len(line.lstrip("#")),
)
else:
if currentElement is not None and currentElement.type != "text":
elements.append(currentElement)
currentElement = Element(
id=f"id_{len(elements)}", type="text", element=line
)
elif currentElement is not None:
currentElement.element += "\n" + line
else:
currentElement = Element(
id=f"id_{len(elements)}", type="text", element=line
)
if currentElement is not None:
elements.append(currentElement)
for idx, element in enumerate(elements):
if element.type == "table":
should_keep = True
perfect_table = True
# verify that the table (markdown) have the same number of columns on each rows
table_lines = element.element.split("\n")
table_columns = [len(line.split("|")) for line in table_lines]
if len(set(table_columns)) > 1:
# if the table have different number of columns on each rows, it's not a perfect table
# we will store the raw text for such tables instead of converting them to a dataframe
perfect_table = False
# verify that the table (markdown) have at least 2 rows
if len(table_lines) < 2:
should_keep = False
# apply the table filter, now only filter empty tables
if should_keep and perfect_table and table_filters is not None:
should_keep = all(tf(element) for tf in table_filters)
# if the element is a table, convert it to a dataframe
if should_keep:
if perfect_table:
table = md_to_df(element.element)
elements[idx] = Element(
id=f"id_{node_id}_{idx}" if node_id else f"id_{idx}",
type="table",
element=element.element,
table=table,
)
else:
# for non-perfect tables, we will store the raw text
# and give it a different type to differentiate it from perfect tables
elements[idx] = Element(
id=f"id_{node_id}_{idx}" if node_id else f"id_{idx}",
type="table_text",
element=element.element,
# table=table
)
else:
elements[idx] = Element(
id=f"id_{node_id}_{idx}" if node_id else f"id_{idx}",
type="text",
element=element.element,
)
else:
# if the element is not a table, keep it as to text
elements[idx] = Element(
id=f"id_{node_id}_{idx}" if node_id else f"id_{idx}",
type="text",
element=element.element,
)
# merge consecutive text elements together for now
merged_elements: List[Element] = []
for element in elements:
if (
len(merged_elements) > 0
and element.type == "text"
and merged_elements[-1].type == "text"
):
merged_elements[-1].element += "\n" + element.element
else:
merged_elements.append(element)
elements = merged_elements
return merged_elements
def filter_table(self, table_element: Any) -> bool:
"""Filter tables."""
table_df = md_to_df(table_element.element)
# check if table_df is not None, has more than one row, and more than one column
return table_df is not None and not table_df.empty and len(table_df.columns) > 1
|