Lexique: work towards word_db

Still problems with other forms

pwgen_fr/lexique.py

@@ -1,12 +1,13 @@
 import csv
 import itertools
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 import logging
 import subprocess
 import typing as t
 from bisect import bisect_left
 import enum
 from pathlib import Path
+from .word_db import Genre, Nombre, Temps, Nom, Adjectif, Verbe, Adverbe, WordDb
 
 logger = logging.getLogger(__name__)
 
@@ -34,84 +35,21 @@ class CatGram(enum.Enum):
         return self.value < oth.value
 
 
-def match_enum_or_all(val, enum_cls) -> list:
+def match_enum_or_all(val: str, enum_mapper, enum_cls) -> list:
     """The value of the enum corresponding if any; else, all terms of the enum"""
-    if val in enum_cls:
-        return [enum_cls(val)]
+    if val in enum_mapper:
+        return [enum_mapper[val]]
     return list(enum_cls)
 
 
-class Genre(enum.Enum):
-    MASC = "m"
-    FEM = "f"
-
-
-class Nombre(enum.Enum):
-    SING = "s"
-    PLUR = "p"
-
-
-class Temps(enum.Enum):
-    INFINITIF = "inf"
-    PRESENT = "ind:pre"
-    FUTUR = "ind:fut"
-    IMPARFAIT = "ind:imp"
-
-
-class Personne(enum.Enum):
-    S1 = "1s"
-    S2 = "2s"
-    S3 = "3s"
-    P1 = "1p"
-    P2 = "2p"
-    P3 = "3p"
-
-
 @dataclass
-class _Mot:
-    """Canonical form of a word"""
-
+class Mot:
     mot: str
+    lemme: str
     cat_gram: CatGram
     freq: float  # occurrences of the canonical form by million words
-
-
-class Mot(_Mot):
-    class Variant:
-        pass
-
-    _for_cat_gram: dict[CatGram, t.Type["Mot"]] = {}
-    _variants: dict
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self._variants = {}
-
-    def accord(self, variant: Variant) -> str:
-        return self._variants[variant]
-
-    @classmethod
-    def for_cat_gram(cls, cat_gram: CatGram) -> t.Type["Mot"]:
-        """The class to use for a word of given CatGram"""
-        return cls._for_cat_gram.get(cat_gram, cls)
-
-
-class Nom(Mot):
-    class Variant(t.NamedTuple):
-        genre: Genre
-        nombre: Nombre
-
-
-class Verbe(Mot):
-    class Variant(t.NamedTuple):
-        temps: Temps
-        personne: t.Optional[Personne]
-
-
-Mot._for_cat_gram = {
-    CatGram.NOM: Nom,
-    CatGram.VERBE: Verbe,
-}
+    variantes: dict[tuple, str] = field(default_factory=dict)
+    genre: t.Optional[Genre] = None
 
 
 class Lexique:
@@ -125,6 +63,32 @@ class Lexique:
         CatGram.ADVERBE: 10000,
     }
 
+    class Parsers:
+        """Datatables to help parse the original data"""
+
+        genre: dict[str, Genre] = {
+            "m": Genre.MASC,
+            "f": Genre.FEM,
+        }
+        rev_genre: dict[t.Optional[Genre], str] = {
+            None: "",
+            Genre.MASC: "m",
+            Genre.FEM: "f",
+        }
+        nombre: dict[str, Nombre] = {
+            "s": Nombre.SING,
+            "p": Nombre.PLUR,
+        }
+        verbe_temps: dict[str, Temps] = {
+            "ind:pre": Temps.PRESENT,
+            "ind:fut": Temps.FUTUR,
+            "ind:imp": Temps.IMPARFAIT,
+        }
+        verbe_personne: dict[str, Nombre] = {
+            "3s": Nombre.SING,
+            "3p": Nombre.PLUR,
+        }
+
     dataset: list[Mot]
 
     def __init__(self, dataset):
@@ -161,6 +125,32 @@ class Lexique:
                 f"Uncompressed dataset still missing at {cls.LEXIQUE_DIR_PATH} after extraction"
             )
 
+    @classmethod
+    def _find_word_key(cls, mot: Mot):
+        return (mot.lemme, mot.cat_gram, cls.Parsers.rev_genre[mot.genre])
+
+    @classmethod
+    def _find_word(cls, dataset: list[Mot], row: dict) -> t.Optional[Mot]:
+        str_lemme = row["lemme"]
+        cat_gram = CatGram.parse(row["cgram"])
+        genre = row["genre"] if cat_gram == CatGram.NOM else ""
+        row_key = (
+            str_lemme,
+            cat_gram,
+            genre,
+        )
+        lemme_pos = bisect_left(
+            dataset,
+            row_key,
+            key=cls._find_word_key,
+        )
+        if lemme_pos >= len(dataset):
+            return None
+        out = dataset[lemme_pos]
+        if row_key != cls._find_word_key(out):
+            return None
+        return dataset[lemme_pos]
+
     @classmethod
     def parse(cls) -> "Lexique":
         out = []
@@ -174,34 +164,38 @@ class Lexique:
 
         # First pass: generate canonical forms (lemmes)
         for row in rows:
-            if row["lemme"] != row["ortho"]:
-                continue
             cat_gram = CatGram.parse(row["cgram"])
+            if (row["lemme"] != row["ortho"]) and not (
+                cat_gram == CatGram.NOM and row["genre"] == "f" and row["nombre"] == "s"
+            ):
+                # Un nom singulier féminin est considéré comme forme canonique
+                continue
+
+            genre: t.Optional[Genre] = None
+            if cat_gram == CatGram.NOM:
+                genre = cls.Parsers.genre.get(row["genre"], None)
             out.append(
-                Mot.for_cat_gram(cat_gram)(
+                Mot(
                     mot=row["ortho"],
+                    lemme=row["lemme"],
                     cat_gram=cat_gram,
                     freq=float(row["freqlemlivres"]),
+                    genre=genre,
                 )
             )
 
-        out.sort(key=lambda x: (x.mot, x.cat_gram))  # We need to bisect on this.
+        out.sort(key=cls._find_word_key)  # We need to bisect on this.
 
         # Second pass: populate variants
         for row in rows:
-            str_lemme = row["lemme"]
-            cat_gram = CatGram.parse(row['cgram'])
-            lemme_pos = bisect_left(out, (str_lemme, cat_gram), key=lambda x: (x.mot, x.cat_gram))
-            if lemme_pos > len(out) or out[lemme_pos].mot != str_lemme:
-                continue  # Unknown word
-            lemme = out[lemme_pos]
+            lemme = cls._find_word(out, row)
+            if lemme is None:
+                continue
 
             if lemme.cat_gram == CatGram.NOM:
-                genres = match_enum_or_all(row["genre"], Genre)
-                nombres = match_enum_or_all(row["nombre"], Nombre)
-                for genre, nombre in itertools.product(genres, nombres):
-                    variant = Nom.Variant(genre=genre, nombre=nombre)
-                    lemme._variants[variant] = row["ortho"]
+                nombres = match_enum_or_all(row["nombre"], cls.Parsers.nombre, Nombre)
+                for nombre in nombres:
+                    lemme.variantes[(nombre,)] = row["ortho"]
 
             elif lemme.cat_gram == CatGram.VERBE:
                 infover = row["infover"].split(";")
@@ -210,20 +204,23 @@ class Lexique:
 
                     temps = None
                     personne = None
-                    if ver[0] == "inf":
-                        temps = Temps(ver[0])
-                    elif ver[0] == "ind":
                     temps_select = ":".join(ver[0:2])
                     if temps_select not in Temps:
                         continue
                     temps = Temps(temps_select)
-                        personne = Personne(ver[2])
-                    else:
-                        continue
+                    personne = cls.Parsers.verbe_personne.get(ver[2], None)
+                    if personne is None:
+                        continue  # we're not interested in all conj. persons
 
-                    variant = Verbe.Variant(temps=temps, personne=personne)
-                    lemme._variants[variant] = row["ortho"]
+                    lemme.variantes[(temps, personne)] = row["ortho"]
 
+            elif lemme.cat_gram == CatGram.ADJECTIF:
+                genres = match_enum_or_all(row["genre"], cls.Parsers.genre, Genre)
+                nombres = match_enum_or_all(row["nombre"], cls.Parsers.nombre, Nombre)
+                for genre, nombre in itertools.product(genres, nombres):
+                    lemme.variantes[(genre, nombre)] = row["ortho"]
+
+            # No need to match adverbs (invariant)
         return cls(out)
 
     def most_common(
@@ -240,3 +237,52 @@ class Lexique:
         out = list(filter(lambda word: word.cat_gram == cat_gram, self.dataset))
         out.sort(key=lambda word: word.freq, reverse=True)
         return out[:threshold]
+
+    def word_db(self, thresholds: t.Optional[dict[CatGram, int]] = None) -> WordDb:
+        """Convert to a WordDb"""
+        thresholds = thresholds or {}
+
+        noms = self.most_common(CatGram.NOM, thresholds.get(CatGram.NOM, None))
+        db_noms = [
+            Nom(
+                genre=t.cast(Genre, nom.genre),  # not None for noms
+                sing=nom.variantes[(Nombre.SING,)],
+                plur=nom.variantes[(Nombre.PLUR,)],
+            )
+            for nom in noms
+        ]
+
+        adjectifs = self.most_common(
+            CatGram.ADJECTIF, thresholds.get(CatGram.ADJECTIF, None)
+        )
+        db_adjectifs = [
+            Adjectif(
+                masc_sing=adj.variantes[(Genre.MASC, Nombre.SING)],
+                masc_plur=adj.variantes[(Genre.MASC, Nombre.PLUR)],
+                fem_sing=adj.variantes[(Genre.FEM, Nombre.SING)],
+                fem_plur=adj.variantes[(Genre.FEM, Nombre.PLUR)],
+            )
+            for adj in adjectifs
+        ]
+
+        verbes = self.most_common(CatGram.VERBE, thresholds.get(CatGram.VERBE, None))
+        db_verbes = [
+            Verbe(
+                present_sing=verbe.variantes[(Temps.PRESENT, Nombre.SING)],
+                present_plur=verbe.variantes[(Temps.PRESENT, Nombre.PLUR)],
+                futur_sing=verbe.variantes[(Temps.FUTUR, Nombre.SING)],
+                futur_plur=verbe.variantes[(Temps.FUTUR, Nombre.PLUR)],
+                imparfait_sing=verbe.variantes[(Temps.IMPARFAIT, Nombre.SING)],
+                imparfait_plur=verbe.variantes[(Temps.IMPARFAIT, Nombre.PLUR)],
+            )
+            for verbe in verbes
+        ]
+
+        adverbes = self.most_common(
+            CatGram.ADVERBE, thresholds.get(CatGram.ADVERBE, None)
+        )
+        db_adverbes = [Adverbe(adv=adv.mot) for adv in adverbes]
+
+        return WordDb(
+            noms=db_noms, adjectifs=db_adjectifs, verbes=db_verbes, adverbes=db_adverbes
+        )

pwgen_fr/word_db.py

@@ -0,0 +1,132 @@
+""" A pre-processed database of words, independant of their source """
+
+import typing as t
+from enum import Enum
+import json
+
+
+class Genre(Enum):
+    MASC = "masculin"
+    FEM = "féminin"
+
+
+class Nombre(Enum):
+    SING = "singulier"
+    PLUR = "pluriel"
+
+
+class Temps(Enum):
+    PRESENT = "present"
+    FUTUR = "futur"
+    IMPARFAIT = "imparfait"
+
+
+class Nom(t.NamedTuple):
+    """Nom commun"""
+
+    genre: Genre
+    sing: str
+    plur: str
+
+    def __str__(self) -> str:
+        return f"{self.sing}"
+
+    def accord(self, nombre: Nombre) -> str:
+        """Accorde en nombre"""
+        return getattr(self, nombre.name.lower())
+
+
+class Adjectif(t.NamedTuple):
+    masc_sing: str
+    masc_plur: str
+    fem_sing: str
+    fem_plur: str
+
+    def __str__(self) -> str:
+        return f"{self.masc_sing}/{self.fem_sing}"
+
+    def accord(self, genre: Genre, nombre: Nombre) -> str:
+        """Accorde en genre et en nombre"""
+        return getattr(self, f"{genre.name.lower()}_{nombre.name.lower()}")
+
+
+class Verbe(t.NamedTuple):
+    present_sing: str
+    present_plur: str
+    futur_sing: str
+    futur_plur: str
+    imparfait_sing: str
+    imparfait_plur: str
+
+    def __str__(self) -> str:
+        return f"{self.present_sing}"
+
+    def accord(self, temps: Temps, nombre: Nombre) -> str:
+        """Accorde en temps et en nombre (seule la 3è pers. est utilisée)"""
+        return getattr(self, f"{temps.name.lower()}_{nombre.name.lower()}")
+
+
+class Adverbe(t.NamedTuple):
+    """Packed as named tuple for consistence"""
+
+    adv: str
+
+    def __str__(self) -> str:
+        return self.adv
+
+    def accord(self) -> str:
+        """for consistence"""
+        return self.adv
+
+
+class WordDb:
+    """Base de donnée de mots, sérialisable"""
+
+    _serialize_data: dict[str, t.Type[t.NamedTuple]] = {
+        "noms": Nom,
+        "adjectifs": Adjectif,
+        "verbes": Verbe,
+        "adverbes": Adverbe,
+    }
+
+    noms: list[Nom]
+    adjectifs: list[Adjectif]
+    verbes: list[Verbe]
+    adverbes: list[Adverbe]
+
+    def __init__(
+        self,
+        noms: t.Optional[list[Nom]] = None,
+        adjectifs: t.Optional[list[Adjectif]] = None,
+        verbes: t.Optional[list[Verbe]] = None,
+        adverbes: t.Optional[list[Adverbe]] = None,
+    ):
+        self.noms = noms or []
+        self.adjectifs = adjectifs or []
+        self.verbes = verbes or []
+        self.adverbes = adverbes or []
+
+    def serialize(self) -> dict:
+        """Serialize to plain dictionary (no classes)"""
+        return {
+            attr: [x._asdict() for x in getattr(self, attr)]
+            for attr in self.__class__._serialize_data
+        }
+
+    def save(self, fd):
+        """Serialize to this stream"""
+        json.dump(self.serialize(), fd)
+
+    @classmethod
+    @t.no_type_check  # serialization is messy
+    def unserialize(cls, data: dict) -> "WordDb":
+        """Reverses :serialize:"""
+        parsed = {}
+        for attr, attr_cls in cls._serialize_data.items():
+            parsed[attr] = list(map(attr_cls, data[attr]))
+        return cls(**parsed)
+
+    @classmethod
+    def load(cls, fd) -> "WordDb":
+        """Unserialize from this stream"""
+        return cls.unserialize(json.load(fd))