| """Chunker behaviour, exercised with a deterministic fake tokenizer.""" |
|
|
| from __future__ import annotations |
|
|
| import pytest |
|
|
| from ehekim.chunking import chunk_article, normalize_text, split_sentences |
|
|
|
|
| class WordTokenizer: |
| """Whitespace tokenizer: one token per word, so counts are predictable.""" |
|
|
| def encode(self, text: str, add_special_tokens: bool = False) -> list[int]: |
| return [hash(w) % 1000 for w in text.split()] |
|
|
| def decode(self, ids, skip_special_tokens: bool = True) -> str: |
| return " ".join("w" for _ in ids) |
|
|
|
|
| @pytest.fixture |
| def tok() -> WordTokenizer: |
| return WordTokenizer() |
|
|
|
|
| class TestNormalize: |
| def test_collapses_excess_blank_lines_and_spaces(self): |
| assert normalize_text("a\n\n\n\nb\r\nc d") == "a\n\nb\nc d" |
|
|
| def test_empty_input(self): |
| assert normalize_text("") == "" |
| assert normalize_text(" \n ") == "" |
|
|
|
|
| class TestSentenceSplitting: |
| def test_splits_on_terminal_punctuation(self): |
| out = split_sentences("Birinci cümle. İkinci cümle! Üçüncü cümle?") |
| assert out == ["Birinci cümle.", "İkinci cümle!", "Üçüncü cümle?"] |
|
|
| @pytest.mark.parametrize( |
| "text", |
| [ |
| "Dr. Ahmet geldi.", |
| "Doz 500 mg. olarak verildi.", |
| "Bunlar vb. durumlardır.", |
| "M. Ali Bey geldi.", |
| ], |
| ) |
| def test_does_not_split_on_abbreviations_or_initials(self, text): |
| assert len(split_sentences(text)) == 1 |
|
|
| def test_no_boundary_returns_whole_text(self): |
| assert split_sentences("tek parça metin") == ["tek parça metin"] |
|
|
|
|
| class TestChunkArticle: |
| def test_short_article_is_one_chunk(self, tok): |
| text = " ".join(f"kelime{i}" for i in range(50)) |
| chunks = chunk_article(text, tok, target_tokens=100, overlap_tokens=10, min_tokens=5) |
| assert len(chunks) == 1 |
| assert chunks[0].index == 0 |
|
|
| def test_respects_the_token_budget(self, tok): |
| paragraphs = ["\n".join([" ".join(f"w{i}" for i in range(40))] * 1) for _ in range(20)] |
| text = "\n".join(paragraphs) |
| chunks = chunk_article(text, tok, target_tokens=100, overlap_tokens=0, min_tokens=1) |
| assert len(chunks) > 1 |
| |
| assert all(c.token_count <= 120 for c in chunks) |
|
|
| def test_single_newlines_are_paragraph_boundaries(self, tok): |
| """The corpus separates paragraphs with one newline, not a blank line.""" |
| text = "\n".join(" ".join(f"p{p}w{i}" for i in range(30)) for p in range(10)) |
| chunks = chunk_article(text, tok, target_tokens=60, overlap_tokens=0, min_tokens=1) |
| assert len(chunks) > 1 |
|
|
| def test_overlap_repeats_content_between_neighbours(self, tok): |
| text = "\n".join(f"paragraf{p} " + " ".join(f"w{i}" for i in range(20)) for p in range(10)) |
| with_overlap = chunk_article(text, tok, target_tokens=60, overlap_tokens=25, min_tokens=1) |
| assert len(with_overlap) >= 2 |
| first_words = set(with_overlap[0].text.split()) |
| second_words = set(with_overlap[1].text.split()) |
| assert first_words & second_words, "ardışık parçalar örtüşmeli" |
|
|
| def test_oversized_single_sentence_is_hard_split(self, tok): |
| text = " ".join(f"w{i}" for i in range(300)) |
| chunks = chunk_article(text, tok, target_tokens=50, overlap_tokens=0, min_tokens=1) |
| assert len(chunks) > 1 |
|
|
| def test_indices_are_sequential_from_zero(self, tok): |
| text = "\n".join(" ".join(f"p{p}w{i}" for i in range(30)) for p in range(12)) |
| chunks = chunk_article(text, tok, target_tokens=60, overlap_tokens=10, min_tokens=1) |
| assert [c.index for c in chunks] == list(range(len(chunks))) |
|
|
| def test_empty_article_yields_nothing(self, tok): |
| assert chunk_article("", tok) == [] |
| assert chunk_article(" \n\n ", tok) == [] |
|
|
