sparrow-parse 0.2.7__tar.gz → 0.2.8__tar.gz

{sparrow_parse-0.2.7 → sparrow_parse-0.2.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sparrow-parse
-Version: 0.2.7
+Version: 0.2.8
 Summary: Sparrow Parse is a Python package for parsing and extracting information from documents.
 Home-page: https://github.com/katanaml/sparrow/tree/main/sparrow-data/parse
 License: GPL-3.0

{sparrow_parse-0.2.7 → sparrow_parse-0.2.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sparrow-parse"
-version = "0.2.7"
+version = "0.2.8"
 description = "Sparrow Parse is a Python package for parsing and extracting information from documents."
 authors = ["Andrej Baranovskij <andrejus.baranovskis@gmail.com>"]
 license = "GPL-3.0"

sparrow_parse-0.2.8/sparrow_parse/__init__.py

@@ -0,0 +1 @@
+__version__ = '0.2.8'

{sparrow_parse-0.2.7 → sparrow_parse-0.2.8}/sparrow_parse/extractor/extractor_helper.py

@@ -3,9 +3,10 @@ from sentence_transformers import SentenceTransformer, util
 import pandas as pd
 import re
 from io import StringIO
+from rich import print
 
 
-def merge_html_table_headers(html_table, column_keywords, debug=False):
+def merge_html_table_headers(html_table, column_keywords, similarity_threshold, debug=False):
     soup = BeautifulSoup(html_table, 'html.parser')
 
     # Find all thead elements
@@ -18,7 +19,7 @@ def merge_html_table_headers(html_table, column_keywords, debug=False):
         html_table = normalize_html_table(html_table, debug)
         html_table = fix_rowspan_elements(html_table)
         html_table = merge_rows_with_rowspan(html_table)
-        html_table = detect_and_remove_junk_columns(html_table, column_keywords, debug)
+        html_table = detect_and_remove_junk_columns(html_table, column_keywords, similarity_threshold, debug)
     else:
         # If there is only one thead, return the original table
         return html_table
@@ -272,7 +273,7 @@ def merge_rows_with_rowspan(html):
     return str(new_table_soup.table)
 
 
-def detect_and_remove_junk_columns(html_table, target_columns, debug=False):
+def detect_and_remove_junk_columns(html_table, target_columns, similarity_threshold_param, debug=False):
     html_table = clean_html_table_header_names(html_table)
 
     # Wrap the HTML string in a StringIO object
@@ -295,7 +296,7 @@ def detect_and_remove_junk_columns(html_table, target_columns, debug=False):
 
     # Identify junk columns based on similarity threshold
     junk_columns = []
-    similarity_threshold = 0.5  # Adjust this threshold as needed
+    similarity_threshold = similarity_threshold_param
 
     for idx, col_embedding in enumerate(column_embeddings):
         similarities = util.pytorch_cos_sim(col_embedding, target_embeddings)[0]
@@ -357,11 +358,11 @@ def clean_html_table_header_names(html_table: str) -> str:
     # Extract the headers and clean them
     headers = table.find_all("th")
     for th in headers:
-        clean_header = re.sub(r"[^a-zA-Z0-9\s]", "", th.get_text())
-        # Check if the cleaned name is empty
-        if not clean_header.strip():
-            clean_header = "-"
-        th.string.replace_with(clean_header)
+        if th.string:
+            # Clean the header
+            clean_header = re.sub(r"[^a-zA-Z0-9\s]", "", th.get_text())
+            # Keep it empty if the cleaned name is empty
+            th.string.replace_with(clean_header.strip() if clean_header.strip() else "")
 
     html_table = str(soup)
 

{sparrow_parse-0.2.7 → sparrow_parse-0.2.8}/sparrow_parse/extractor/html_extractor.py

@@ -12,8 +12,8 @@ class HTMLExtractor(object):
     def __init__(self):
         pass
 
-    def read_data(self, target_columns, data, column_keywords=None, group_by_rows=True, update_targets=False,
-                  local=True, debug=False):
+    def read_data(self, target_columns, data, similarity_threshold_junk, similarity_threshold_column_id,
+                  column_keywords=None, group_by_rows=True, update_targets=False, local=True, debug=False):
         answer = {}
 
         json_result, targets_unprocessed = [], []
@@ -22,7 +22,8 @@ class HTMLExtractor(object):
             if not target_columns:
                 break
 
-            json_result, targets_unprocessed = self.read_data_from_table(target_columns, table, column_keywords,
+            json_result, targets_unprocessed = self.read_data_from_table(target_columns, table, similarity_threshold_junk,
+                                                                         similarity_threshold_column_id, column_keywords,
                                                                          group_by_rows, local, debug)
             answer = self.add_answer_section(answer, "items" + str(i + 1), json_result)
 
@@ -33,9 +34,10 @@ class HTMLExtractor(object):
 
         return answer, targets_unprocessed
 
-    def read_data_from_table(self, target_columns, data, column_keywords=None, group_by_rows=True, local=True, debug=False):
+    def read_data_from_table(self, target_columns, data, similarity_threshold_junk, similarity_threshold_column_id,
+                             column_keywords=None, group_by_rows=True, local=True, debug=False):
         data = self.invoke_pipeline_step(
-            lambda: merge_html_table_headers(data, column_keywords, debug),
+            lambda: merge_html_table_headers(data, column_keywords, similarity_threshold_junk, debug),
             "Merging HTML table headers...",
             local
         )
@@ -54,7 +56,7 @@ class HTMLExtractor(object):
             print(f"Target columns: {target_columns}")
 
         indices, targets, targets_unprocessed = self.invoke_pipeline_step(
-            lambda: self.calculate_similarity(columns, target_columns, debug),
+            lambda: self.calculate_similarity(columns, target_columns, similarity_threshold_column_id, debug),
             "Calculating cosine similarity between columns and target values...",
             local
         )
@@ -73,7 +75,7 @@ class HTMLExtractor(object):
 
         return json_result, targets_unprocessed
 
-    def calculate_similarity(self, columns, target_columns, debug):
+    def calculate_similarity(self, columns, target_columns, similarity_threshold_column_id, debug):
         model = SentenceTransformer('all-mpnet-base-v2')
 
         # Compute embeddings for columns and target values
@@ -93,7 +95,7 @@ class HTMLExtractor(object):
             most_similar_idx = similarities.argmax().item()
             most_similar_column = columns[most_similar_idx]
             similarity_score = similarities[most_similar_idx].item()
-            if similarity_score > 0.3:
+            if similarity_score > similarity_threshold_column_id:
                 if most_similar_idx in most_similar_indices:
                     if similarity_score > most_similar_indices[most_similar_idx][1]:
                         targets_unprocessed.append(most_similar_indices[most_similar_idx][0])
@@ -222,27 +224,29 @@ if __name__ == "__main__":
     # to run for debugging, navigate to sparrow_parse and run the following command:
     # python -m extractor.html_extractor
 
-    # with open('data/invoice_1_table.txt', 'r') as file:
-    #     file_content = file.read()
-    #
-    # file_content = file_content.strip()[1:-1].strip()
-    # data_list = re.split(r"',\s*'", file_content)
-    # data_list = [item.strip(" '") for item in data_list]
+    with open('data/invoice_1_table.txt', 'r') as file:
+        file_content = file.read()
+
+    file_content = file_content.strip()[1:-1].strip()
+    data_list = re.split(r"',\s*'", file_content)
+    data_list = [item.strip(" '") for item in data_list]
 
     extractor = HTMLExtractor()
 
-    # answer, targets_unprocessed = extractor.read_data(
-    #     ['description', 'qty', 'net_price', 'net_worth', 'vat', 'gross_worth'],
-    #     # ['transaction_date', 'value_date', 'description', 'cheque', 'withdrawal', 'deposit', 'balance',
-    #     #  'deposits', 'account_number', 'od_limit', 'currency_balance', 'sgd_balance', 'maturity_date'],
-    #     data_list,
-    #     None,
-    #     # ['deposits', 'account_number', 'od_limit', 'currency_balance', 'sgd_balance', 'transaction_date',
-    #     #  'value_date', 'description', 'cheque', 'withdrawal', 'deposit', 'balance', 'maturity_date'],
-    #     True,
-    #     True,
-    #     True,
-    #     True)
-    #
-    # print(answer)
-    # print(targets_unprocessed)
+    answer, targets_unprocessed = extractor.read_data(
+        # ['description', 'qty', 'net_price', 'net_worth', 'vat', 'gross_worth'],
+        ['transaction_date', 'value_date', 'description', 'cheque', 'withdrawal', 'deposit', 'balance',
+         'deposits', 'account_number', 'od_limit', 'currency_balance', 'sgd_balance', 'maturity_date'],
+        data_list,
+        0.5,
+        0.3,
+        # None,
+        ['deposits', 'account_number', 'od_limit', 'currency_balance', 'sgd_balance', 'transaction_date',
+         'value_date', 'description', 'cheque', 'withdrawal', 'deposit', 'balance', 'maturity_date'],
+        True,
+        True,
+        True,
+        True)
+
+    print(answer)
+    print(targets_unprocessed)

{sparrow_parse-0.2.7 → sparrow_parse-0.2.8}/sparrow_parse/extractor/unstructured_processor.py

@@ -177,3 +177,11 @@ if __name__ == "__main__":
     #     True,
     #     True)
 
+    content, table_content = processor.extract_data(
+        '/Users/andrejb/Documents/work/epik/bankstatement/OCBC_1_1.pdf',
+        'hi_res',
+        'yolox',
+        ['tables', 'unstructured'],
+        True,
+        True)
+

sparrow_parse-0.2.7/sparrow_parse/__init__.py

@@ -1 +0,0 @@
-__version__ = '0.2.7'