Add maximum_length argument to chunk_sentences

cognee/tasks/chunks/chunk_by_paragraph.py

@@ -13,7 +13,8 @@ def chunk_by_paragraph(data: str, paragraph_length: int = 1024, batch_paragraphs
     last_paragraph_id = None
     last_cut_type = None
     
-    for paragraph_id, _, sentence, word_count, end_type in chunk_by_sentence(data):
+    for paragraph_id, _, sentence, word_count, end_type in chunk_by_sentence(data, maximum_length=paragraph_length):
+        assert word_count <= paragraph_length, f"{paragraph_length = } is smaller than {word_count = }"
         # Check if this sentence would exceed length limit
         if current_word_count > 0 and current_word_count + word_count > paragraph_length:
             # Yield current chunk

cognee/tasks/chunks/chunk_by_sentence.py

@@ -2,9 +2,10 @@
 
 
 from uuid import uuid4
+from typing import Optional
 from .chunk_by_word import chunk_by_word
 
-def chunk_by_sentence(data: str):
+def chunk_by_sentence(data: str, maximum_length: Optional[int]):
     sentence = ""
     paragraph_id = uuid4()
     chunk_index = 0
@@ -14,7 +15,7 @@ def chunk_by_sentence(data: str):
         sentence += word
         word_count += 1
 
-        if word_type == "paragraph_end" or word_type == "sentence_end":
+        if word_type == "paragraph_end" or word_type == "sentence_end" or ((word_count is not None) and (word_count == maximum_length)):
             yield (paragraph_id, chunk_index, sentence, word_count, word_type)
             sentence = ""
             word_count = 0

cognee/tests/unit/processing/chunks/chunk_by_paragraph_test2.py

@@ -1,7 +1,7 @@
 import numpy as np
 import pytest
 
-from cognee.tasks.chunks import chunk_by_paragraph
+from cognee.tasks.chunks import chunk_by_paragraph, chunk_by_word
 from cognee.tests.unit.processing.chunks.test_input import INPUT_TEXTS
 
 
@@ -40,3 +40,45 @@ def test_chunk_by_paragraph_isomorphism(input_text, paragraph_length, batch_para
     assert (
         reconstructed_text == input_text
     ), f"texts are not identical: {len(input_text) = }, {len(reconstructed_text) = }"
+
+
+@pytest.mark.parametrize(
+    "input_text,paragraph_length,batch_paragraphs",
+    [
+        (INPUT_TEXTS["english_text"], 64, True),
+        (INPUT_TEXTS["english_text"], 64, False),
+        (INPUT_TEXTS["english_text"], 256, True),
+        (INPUT_TEXTS["english_text"], 256, False),
+        (INPUT_TEXTS["english_text"], 1024, True),
+        (INPUT_TEXTS["english_text"], 1024, False),
+        (INPUT_TEXTS["english_lists"], 64, True),
+        (INPUT_TEXTS["english_lists"], 64, False),
+        (INPUT_TEXTS["english_lists"], 256, True),
+        (INPUT_TEXTS["english_lists"], 256, False),
+        (INPUT_TEXTS["english_lists"], 1024, True),
+        (INPUT_TEXTS["english_lists"], 1024, False),
+        (INPUT_TEXTS["python_code"], 64, True),
+        (INPUT_TEXTS["python_code"], 64, False),
+        (INPUT_TEXTS["python_code"], 256, True),
+        (INPUT_TEXTS["python_code"], 256, False),
+        (INPUT_TEXTS["python_code"], 1024, True),
+        (INPUT_TEXTS["python_code"], 1024, False),
+        (INPUT_TEXTS["chinese_text"], 64, True),
+        (INPUT_TEXTS["chinese_text"], 64, False),
+        (INPUT_TEXTS["chinese_text"], 256, True),
+        (INPUT_TEXTS["chinese_text"], 256, False),
+        (INPUT_TEXTS["chinese_text"], 1024, True),
+        (INPUT_TEXTS["chinese_text"], 1024, False),
+    ],
+)
+def test_paragraph_chunk_length(input_text, paragraph_length, batch_paragraphs):
+    chunks = list(chunk_by_paragraph(input_text, paragraph_length, batch_paragraphs))
+
+    chunk_lengths = np.array(
+        [len(list(chunk_by_word(chunk["text"]))) for chunk in chunks]
+    )
+
+    larger_chunks = chunk_lengths[chunk_lengths > paragraph_length]
+    assert np.all(
+        chunk_lengths <= paragraph_length
+    ), f"{paragraph_length = }: {larger_chunks} are too large"