glitchlings 0.2.6__cp310-cp310-macosx_11_0_universal2.whl → 0.4.0__cp310-cp310-macosx_11_0_universal2.whl

glitchlings/zoo/adjax.py

@@ -0,0 +1,131 @@
+from __future__ import annotations
+
+import random
+from typing import Any
+
+from ._rate import resolve_rate
+from ._text_utils import split_preserving_whitespace, split_token_edges
+from .core import AttackWave, Glitchling
+
+try:
+    from glitchlings._zoo_rust import swap_adjacent_words as _swap_adjacent_words_rust
+except ImportError:  # pragma: no cover - optional acceleration
+    _swap_adjacent_words_rust = None
+
+
+def _python_swap_adjacent_words(
+    text: str,
+    *,
+    rate: float,
+    rng: random.Random,
+) -> str:
+    """Swap the cores of adjacent words while keeping affixes and spacing intact."""
+
+    tokens = split_preserving_whitespace(text)
+    if len(tokens) < 2:
+        return text
+
+    word_indices: list[int] = []
+    for index in range(len(tokens)):
+        token = tokens[index]
+        if not token or token.isspace():
+            continue
+        if index % 2 == 0:
+            word_indices.append(index)
+
+    if len(word_indices) < 2:
+        return text
+
+    clamped = max(0.0, min(rate, 1.0))
+    if clamped <= 0.0:
+        return text
+
+    for cursor in range(0, len(word_indices) - 1, 2):
+        left_index = word_indices[cursor]
+        right_index = word_indices[cursor + 1]
+
+        left_token = tokens[left_index]
+        right_token = tokens[right_index]
+
+        left_prefix, left_core, left_suffix = split_token_edges(left_token)
+        right_prefix, right_core, right_suffix = split_token_edges(right_token)
+
+        if not left_core or not right_core:
+            continue
+
+        should_swap = clamped >= 1.0 or rng.random() < clamped
+        if not should_swap:
+            continue
+
+        tokens[left_index] = f"{left_prefix}{right_core}{left_suffix}"
+        tokens[right_index] = f"{right_prefix}{left_core}{right_suffix}"
+
+    return "".join(tokens)
+
+
+def swap_adjacent_words(
+    text: str,
+    rate: float | None = None,
+    seed: int | None = None,
+    rng: random.Random | None = None,
+    *,
+    swap_rate: float | None = None,
+) -> str:
+    """Swap adjacent word cores while preserving spacing and punctuation."""
+
+    effective_rate = resolve_rate(
+        rate=rate,
+        legacy_value=swap_rate,
+        default=0.5,
+        legacy_name="swap_rate",
+    )
+    clamped_rate = max(0.0, min(effective_rate, 1.0))
+
+    if rng is None:
+        rng = random.Random(seed)
+
+    if _swap_adjacent_words_rust is not None:
+        return _swap_adjacent_words_rust(text, clamped_rate, rng)
+
+    return _python_swap_adjacent_words(text, rate=clamped_rate, rng=rng)
+
+
+class Adjax(Glitchling):
+    """Glitchling that swaps adjacent words to scramble local semantics."""
+
+    def __init__(
+        self,
+        *,
+        rate: float | None = None,
+        swap_rate: float | None = None,
+        seed: int | None = None,
+    ) -> None:
+        self._param_aliases = {"swap_rate": "rate"}
+        effective_rate = resolve_rate(
+            rate=rate,
+            legacy_value=swap_rate,
+            default=0.5,
+            legacy_name="swap_rate",
+        )
+        super().__init__(
+            name="Adjax",
+            corruption_function=swap_adjacent_words,
+            scope=AttackWave.WORD,
+            seed=seed,
+            rate=effective_rate,
+        )
+
+    def pipeline_operation(self) -> dict[str, Any] | None:
+        rate = self.kwargs.get("rate")
+        if rate is None:
+            return None
+        return {
+            "type": "swap_adjacent",
+            "swap_rate": float(rate),
+        }
+
+
+adjax = Adjax()
+
+
+__all__ = ["Adjax", "adjax", "swap_adjacent_words"]

glitchlings/zoo/core.py

@@ -59,18 +59,26 @@ else:
         def with_transform(self, function: Any) -> "Dataset": ...
 
 
-def _is_transcript(value: Any) -> bool:
-    """Return True when the value resembles a chat transcript."""
+def _is_transcript(
+    value: Any,
+    *,
+    allow_empty: bool = True,
+    require_all_content: bool = False,
+) -> bool:
+    """Return `True` when `value` appears to be a chat transcript."""
 
     if not isinstance(value, list):
         return False
 
     if not value:
-        return True
+        return allow_empty
 
     if not all(isinstance(turn, dict) for turn in value):
         return False
 
+    if require_all_content:
+        return all("content" in turn for turn in value)
+
     return "content" in value[-1]
 
 
@@ -233,21 +241,15 @@ class Glitchling:
             message = "datasets is not installed"
             raise ModuleNotFoundError(message) from _datasets_error
 
-        def _is_transcript(value: Any) -> bool:
-            """Return ``True`` when the value resembles a chat transcript."""
-
-            if not isinstance(value, list) or not value:
-                return False
-
-            return all(
-                isinstance(turn, dict) and "content" in turn for turn in value
-            )
-
         def __corrupt_row(row: dict[str, Any]) -> dict[str, Any]:
             row = dict(row)
             for column in columns:
                 value = row[column]
-                if _is_transcript(value):
+                if _is_transcript(
+                    value,
+                    allow_empty=False,
+                    require_all_content=True,
+                ):
                     row[column] = self.corrupt(value)
                 elif isinstance(value, list):
                     row[column] = [self.corrupt(item) for item in value]

glitchlings/zoo/jargoyle.py

@@ -2,121 +2,47 @@ import random
 import re
 from collections.abc import Iterable
 from dataclasses import dataclass
-from typing import TYPE_CHECKING, Any, Literal, cast
-
-try:  # pragma: no cover - exercised in environments with NLTK installed
-    import nltk  # type: ignore[import]
-except ModuleNotFoundError as exc:  # pragma: no cover - triggered when NLTK missing
-    nltk = None  # type: ignore[assignment]
-    find = None  # type: ignore[assignment]
-    _NLTK_IMPORT_ERROR = exc
-else:  # pragma: no cover - executed when NLTK is available
-    from nltk.corpus.reader import WordNetCorpusReader as _WordNetCorpusReader  # type: ignore[import]
-    from nltk.data import find as _nltk_find  # type: ignore[import]
-
-    find = _nltk_find
-    _NLTK_IMPORT_ERROR = None
-
-if TYPE_CHECKING:  # pragma: no cover - typing aid only
-    from nltk.corpus.reader import WordNetCorpusReader  # type: ignore[import]
-else:  # Use ``Any`` at runtime to avoid hard dependency when NLTK missing
-    WordNetCorpusReader = Any
-
-if nltk is not None:  # pragma: no cover - guarded by import success
-    try:
-        from nltk.corpus import wordnet as _WORDNET_MODULE  # type: ignore[import]
-    except ModuleNotFoundError:  # pragma: no cover - only hit on namespace packages
-        _WORDNET_MODULE = None
-    else:
-        WordNetCorpusReader = _WordNetCorpusReader  # type: ignore[assignment]
-else:
-    _WORDNET_MODULE = None
+from typing import Any, Literal, cast
 
-from .core import AttackWave, Glitchling
-from ._rate import resolve_rate
-
-_WORDNET_HANDLE: WordNetCorpusReader | Any | None = _WORDNET_MODULE
-
-_wordnet_ready = False
+from glitchlings.lexicon import Lexicon, get_default_lexicon
 
+try:  # pragma: no cover - optional WordNet dependency
+    from glitchlings.lexicon.wordnet import (
+        WordNetLexicon,
+        dependencies_available as _lexicon_dependencies_available,
+        ensure_wordnet as _lexicon_ensure_wordnet,
+    )
+except Exception:  # pragma: no cover - triggered when nltk unavailable
+    WordNetLexicon = None  # type: ignore[assignment]
 
-def _require_nltk() -> None:
-    """Ensure the NLTK dependency is present before continuing."""
+    def _lexicon_dependencies_available() -> bool:
+        return False
 
-    if nltk is None or find is None:
-        message = (
-            "The NLTK package is required for the jargoyle glitchling; install "
-            "the 'wordnet' extra via `pip install glitchlings[wordnet]`."
+    def _lexicon_ensure_wordnet() -> None:
+        raise RuntimeError(
+            "The WordNet backend is no longer bundled by default. Install NLTK "
+            "and download its WordNet corpus manually if you need legacy synonyms."
         )
-        if '_NLTK_IMPORT_ERROR' in globals() and _NLTK_IMPORT_ERROR is not None:
-            raise RuntimeError(message) from _NLTK_IMPORT_ERROR
-        raise RuntimeError(message)
-
-
-def dependencies_available() -> bool:
-    """Return ``True`` when the runtime NLTK dependency is present."""
-
-    return nltk is not None and find is not None
-
-
-def _load_wordnet_reader() -> WordNetCorpusReader:
-    """Return a WordNet corpus reader from the downloaded corpus files."""
 
-    _require_nltk()
-
-    try:
-        root = find("corpora/wordnet")
-    except LookupError:
-        try:
-            zip_root = find("corpora/wordnet.zip")
-        except LookupError as exc:
-            raise RuntimeError(
-                "The NLTK WordNet corpus is not installed; run `nltk.download('wordnet')`."
-            ) from exc
-        root = zip_root.join("wordnet/")
-
-    return WordNetCorpusReader(root, None)
-
-
-def _wordnet(force_refresh: bool = False) -> WordNetCorpusReader | Any:
-    """Retrieve the active WordNet handle, rebuilding it on demand."""
-
-    global _WORDNET_HANDLE
-
-    if force_refresh:
-        _WORDNET_HANDLE = _WORDNET_MODULE
-
-    if _WORDNET_HANDLE is not None:
-        return _WORDNET_HANDLE
-
-    _WORDNET_HANDLE = _load_wordnet_reader()
-    return _WORDNET_HANDLE
 
+from ._rate import resolve_rate
+from .core import AttackWave, Glitchling
 
-def ensure_wordnet() -> None:
-    """Ensure the WordNet corpus is available before use."""
+ensure_wordnet = _lexicon_ensure_wordnet
 
-    global _wordnet_ready
-    if _wordnet_ready:
-        return
 
-    _require_nltk()
+def dependencies_available() -> bool:
+    """Return ``True`` when a synonym backend is accessible."""
 
-    resource = _wordnet()
+    if _lexicon_dependencies_available():
+        return True
 
     try:
-        resource.ensure_loaded()
-    except LookupError:
-        nltk.download("wordnet", quiet=True)
-        try:
-            resource = _wordnet(force_refresh=True)
-            resource.ensure_loaded()
-        except LookupError as exc:  # pragma: no cover - only triggered when download fails
-            raise RuntimeError(
-                "Unable to load NLTK WordNet corpus for the jargoyle glitchling."
-            ) from exc
-
-    _wordnet_ready = True
+        # Fall back to the configured default lexicon (typically the bundled vector cache).
+        get_default_lexicon(seed=None)
+    except Exception:
+        return False
+    return True
 
 
 # Backwards compatibility for callers relying on the previous private helper name.
@@ -140,7 +66,9 @@ def _split_token(token: str) -> tuple[str, str, str]:
     return prefix, core, suffix
 
 
-def _normalize_parts_of_speech(part_of_speech: PartOfSpeechInput) -> NormalizedPartsOfSpeech:
+def _normalize_parts_of_speech(
+    part_of_speech: PartOfSpeechInput,
+) -> NormalizedPartsOfSpeech:
     """Coerce user input into a tuple of valid WordNet POS tags."""
 
     if isinstance(part_of_speech, str):
@@ -173,41 +101,8 @@ class CandidateInfo:
     prefix: str
     core_word: str
     suffix: str
-    parts_of_speech: NormalizedPartsOfSpeech
-
-
-def _collect_synonyms(
-    word: str, parts_of_speech: NormalizedPartsOfSpeech
-) -> list[str]:
-    """Gather deterministic synonym candidates for the supplied word."""
-
-    normalized_word = word.lower()
-    wordnet = _wordnet()
-    synonyms: set[str] = set()
-    for pos_tag in parts_of_speech:
-        synsets = wordnet.synsets(word, pos=pos_tag)
-        if not synsets:
-            continue
-
-        for synset in synsets:
-            lemmas_list = [lemma.name() for lemma in cast(Any, synset).lemmas()]
-            if not lemmas_list:
-                continue
-
-            filtered = []
-            for lemma_str in lemmas_list:
-                cleaned = lemma_str.replace("_", " ")
-                if cleaned.lower() != normalized_word:
-                    filtered.append(cleaned)
-
-            if filtered:
-                synonyms.update(filtered)
-                break
-
-        if synonyms:
-            break
-
-    return sorted(synonyms)
+    part_of_speech: str | None
+    synonyms: list[str]
 
 
 def substitute_random_synonyms(
@@ -218,22 +113,27 @@ def substitute_random_synonyms(
     rng: random.Random | None = None,
     *,
     replacement_rate: float | None = None,
+    lexicon: Lexicon | None = None,
 ) -> str:
-    """Replace words with random WordNet synonyms.
+    """Replace words with random lexicon-driven synonyms.
 
     Parameters
     - text: Input text.
-    - rate: Max proportion of candidate words to replace (default 0.1).
+    - rate: Max proportion of candidate words to replace (default 0.01).
     - part_of_speech: WordNet POS tag(s) to target. Accepts "n", "v", "a", "r",
-      any iterable of those tags, or "any" to include all four.
+      any iterable of those tags, or "any" to include all four. Backends that do
+      not differentiate parts of speech simply ignore the setting.
     - rng: Optional RNG instance used for deterministic sampling.
     - seed: Optional seed if `rng` not provided.
+    - lexicon: Optional :class:`~glitchlings.lexicon.Lexicon` implementation to
+      supply synonyms. Defaults to the configured lexicon priority, typically the
+      packaged vector cache.
 
     Determinism
     - Candidates collected in left-to-right order; no set() reordering.
     - Replacement positions chosen via rng.sample.
-    - Synonyms sorted before rng.choice to fix ordering.
-    - For each POS, the first synset containing alternate lemmas is used for stability.
+    - Synonyms sourced through the lexicon; the default backend derives
+      deterministic subsets per word and part-of-speech using the active seed.
     """
     effective_rate = resolve_rate(
         rate=rate,
@@ -242,68 +142,106 @@ def substitute_random_synonyms(
         legacy_name="replacement_rate",
     )
 
-    ensure_wordnet()
-    wordnet = _wordnet()
-
     active_rng: random.Random
     if rng is not None:
         active_rng = rng
     else:
         active_rng = random.Random(seed)
 
-    target_pos = _normalize_parts_of_speech(part_of_speech)
+    active_lexicon: Lexicon
+    restore_lexicon_seed = False
+    original_lexicon_seed: int | None = None
 
-    # Split but keep whitespace separators so we can rebuild easily
-    tokens = re.split(r"(\s+)", text)
+    if lexicon is None:
+        active_lexicon = get_default_lexicon(seed=seed)
+    else:
+        active_lexicon = lexicon
+        if seed is not None:
+            original_lexicon_seed = active_lexicon.seed
+            if original_lexicon_seed != seed:
+                active_lexicon.reseed(seed)
+                restore_lexicon_seed = True
 
-    # Collect indices of candidate tokens (even positions 0,2,.. are words given our split design)
-    candidate_indices: list[int] = []
-    candidate_metadata: dict[int, CandidateInfo] = {}
-    for idx, tok in enumerate(tokens):
-        if idx % 2 == 0 and tok and not tok.isspace():
-            prefix, core_word, suffix = _split_token(tok)
-            if not core_word:
+    try:
+        target_pos = _normalize_parts_of_speech(part_of_speech)
+
+        # Split but keep whitespace separators so we can rebuild easily
+        tokens = re.split(r"(\s+)", text)
+
+        # Collect indices of candidate tokens (even positions 0,2,.. are words given our split design)
+        candidate_indices: list[int] = []
+        candidate_metadata: dict[int, CandidateInfo] = {}
+        for idx, tok in enumerate(tokens):
+            if idx % 2 == 0 and tok and not tok.isspace():
+                prefix, core_word, suffix = _split_token(tok)
+                if not core_word:
+                    continue
+
+                chosen_pos: str | None = None
+                synonyms: list[str] = []
+
+                for pos in target_pos:
+                    if not active_lexicon.supports_pos(pos):
+                        continue
+                    synonyms = active_lexicon.get_synonyms(core_word, pos=pos)
+                    if synonyms:
+                        chosen_pos = pos
+                        break
+
+                if not synonyms and active_lexicon.supports_pos(None):
+                    synonyms = active_lexicon.get_synonyms(core_word, pos=None)
+
+                if synonyms:
+                    candidate_indices.append(idx)
+                    candidate_metadata[idx] = CandidateInfo(
+                        prefix=prefix,
+                        core_word=core_word,
+                        suffix=suffix,
+                        part_of_speech=chosen_pos,
+                        synonyms=synonyms,
+                    )
+
+        if not candidate_indices:
+            return text
+
+        clamped_rate = max(0.0, effective_rate)
+        if clamped_rate == 0.0:
+            return text
+
+        population = len(candidate_indices)
+        effective_fraction = min(clamped_rate, 1.0)
+        expected_replacements = population * effective_fraction
+        max_replacements = int(expected_replacements)
+        remainder = expected_replacements - max_replacements
+        if remainder > 0.0 and active_rng.random() < remainder:
+            max_replacements += 1
+        if clamped_rate >= 1.0:
+            max_replacements = population
+        max_replacements = min(population, max_replacements)
+        if max_replacements <= 0:
+            return text
+
+        # Choose which positions to replace deterministically via rng.sample
+        replace_positions = active_rng.sample(candidate_indices, k=max_replacements)
+        # Process in ascending order to avoid affecting later indices
+        replace_positions.sort()
+
+        for pos in replace_positions:
+            metadata = candidate_metadata[pos]
+            if not metadata.synonyms:
                 continue
 
-            available_pos: NormalizedPartsOfSpeech = tuple(
-                pos for pos in target_pos if wordnet.synsets(core_word, pos=pos)
-            )
-            if available_pos:
-                candidate_indices.append(idx)
-                candidate_metadata[idx] = CandidateInfo(
-                    prefix=prefix,
-                    core_word=core_word,
-                    suffix=suffix,
-                    parts_of_speech=available_pos,
-                )
-
-    if not candidate_indices:
-        return text
-
-    clamped_rate = max(0.0, effective_rate)
-    max_replacements = int(len(candidate_indices) * clamped_rate)
-    if max_replacements <= 0:
-        return text
-
-    # Choose which positions to replace deterministically via rng.sample
-    replace_positions = active_rng.sample(candidate_indices, k=max_replacements)
-    # Process in ascending order to avoid affecting later indices
-    replace_positions.sort()
+            replacement = active_rng.choice(metadata.synonyms)
+            tokens[pos] = f"{metadata.prefix}{replacement}{metadata.suffix}"
 
-    for pos in replace_positions:
-        metadata = candidate_metadata[pos]
-        synonyms = _collect_synonyms(metadata.core_word, metadata.parts_of_speech)
-        if not synonyms:
-            continue
-
-        replacement = active_rng.choice(synonyms)
-        tokens[pos] = f"{metadata.prefix}{replacement}{metadata.suffix}"
-
-    return "".join(tokens)
+        return "".join(tokens)
+    finally:
+        if restore_lexicon_seed:
+            active_lexicon.reseed(original_lexicon_seed)
 
 
 class Jargoyle(Glitchling):
-    """Glitchling that swaps words with random WordNet synonyms."""
+    """Glitchling that swaps words with lexicon-driven synonyms."""
 
     def __init__(
         self,
@@ -312,22 +250,74 @@ class Jargoyle(Glitchling):
         replacement_rate: float | None = None,
         part_of_speech: PartOfSpeechInput = "n",
         seed: int | None = None,
+        lexicon: Lexicon | None = None,
     ) -> None:
         self._param_aliases = {"replacement_rate": "rate"}
+        self._owns_lexicon = lexicon is None
+        self._external_lexicon_original_seed = (
+            lexicon.seed if isinstance(lexicon, Lexicon) else None
+        )
+        self._initializing = True
         effective_rate = resolve_rate(
             rate=rate,
             legacy_value=replacement_rate,
-            default=0.1,
+            default=0.01,
             legacy_name="replacement_rate",
         )
-        super().__init__(
-            name="Jargoyle",
-            corruption_function=substitute_random_synonyms,
-            scope=AttackWave.WORD,
-            seed=seed,
-            rate=effective_rate,
-            part_of_speech=part_of_speech,
-        )
+        prepared_lexicon = lexicon or get_default_lexicon(seed=seed)
+        if lexicon is not None and seed is not None:
+            prepared_lexicon.reseed(seed)
+        try:
+            super().__init__(
+                name="Jargoyle",
+                corruption_function=substitute_random_synonyms,
+                scope=AttackWave.WORD,
+                seed=seed,
+                rate=effective_rate,
+                part_of_speech=part_of_speech,
+                lexicon=prepared_lexicon,
+            )
+        finally:
+            self._initializing = False
+
+    def set_param(self, key: str, value: Any) -> None:
+        super().set_param(key, value)
+
+        aliases = getattr(self, "_param_aliases", {})
+        canonical = aliases.get(key, key)
+
+        if canonical == "seed":
+            current_lexicon = getattr(self, "lexicon", None)
+            if isinstance(current_lexicon, Lexicon):
+                if getattr(self, "_owns_lexicon", False):
+                    current_lexicon.reseed(self.seed)
+                else:
+                    if self.seed is not None:
+                        current_lexicon.reseed(self.seed)
+                    else:
+                        if hasattr(self, "_external_lexicon_original_seed"):
+                            original_seed = getattr(
+                                self, "_external_lexicon_original_seed", None
+                            )
+                            current_lexicon.reseed(original_seed)
+        elif canonical == "lexicon" and isinstance(value, Lexicon):
+            if getattr(self, "_initializing", False):
+                if getattr(self, "_owns_lexicon", False):
+                    if self.seed is not None:
+                        value.reseed(self.seed)
+                else:
+                    if getattr(self, "_external_lexicon_original_seed", None) is None:
+                        self._external_lexicon_original_seed = value.seed
+                    if self.seed is not None:
+                        value.reseed(self.seed)
+                return
+
+            self._owns_lexicon = False
+            self._external_lexicon_original_seed = value.seed
+            if self.seed is not None:
+                value.reseed(self.seed)
+            elif value.seed != self._external_lexicon_original_seed:
+                value.reseed(self._external_lexicon_original_seed)
 
 
 jargoyle = Jargoyle()