improved pdf_converter (#317)

* improved pdf_converter * fix test
cdqa-suite · Dec 2, 2019 · 5d2e4b3 · 5d2e4b3
1 parent 424e450
commit 5d2e4b3
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 7 deletions.
diff --git a/cdqa/utils/converters.py b/cdqa/utils/converters.py
@@ -170,8 +170,8 @@ def pdf_converter(directory_path, min_length=200, include_line_breaks=False):
         try:
             df.loc[i] = [pdf.replace(".pdf",''), None]
             raw = parser.from_file(os.path.join(directory_path, pdf))
-            s = raw["content"]
-            paragraphs = re.split("\n(?=\u2028|[A-Z-0-9])", s)
+            s = raw["content"].strip()
+            paragraphs = re.split("\n\n(?=\u2028|[A-Z-0-9])", s)
             list_par = []
             temp_para = ""  # variable that stores paragraphs with length<min_length
             # (considered as a line)
@@ -198,9 +198,9 @@ def pdf_converter(directory_path, min_length=200, include_line_breaks=False):
                     else:
                         # appending paragraph p as is to list_par
                         list_par.append(p.replace("\n", ""))
-            else:
-                if temp_para:
-                    list_par.append(temp_para.strip())
+                else:
+                    if temp_para:
+                        list_par.append(temp_para.strip())
 
             df.loc[i, "paragraphs"] = list_par
         except:

diff --git a/setup.py b/setup.py
@@ -8,7 +8,7 @@ def read(file):
 
 setup(
     name="cdqa",
-    version="1.3.8",
+    version="1.3.9",
     author="Félix MIKAELIAN, André FARIAS, Matyas AMROUCHE, Olivier SANS, Théo NAZON",
     description="An End-To-End Closed Domain Question Answering System",
     long_description=read("README.md"),

diff --git a/tests/test_converters.py b/tests/test_converters.py
@@ -39,7 +39,7 @@ def df_converter_check(self, df, include_line_breaks=False):
         if include_line_breaks:
             para_len = [len(df.paragraphs[i]) for i in range(df.shape[0])]
             para_len.sort()
-            if not para_len == [144, 220, 265]:
+            if not para_len == [58, 80, 87]:
                 errors.append(f"error in number of paragraphs : {para_len}")
 
         # assert no error message has been registered, else print messages