Skip to content

Embeddings

kenon.embeddings.CountVectorizerEmbedder

Corpus-internal count-based token embeddings via sklearn CountVectorizer.

Parameters:

Name Type Description Default
stopwords frozenset[str] | None

Stopword set to exclude. Pass None to keep all tokens.

 None
min_df int | float

Minimum document frequency (int or float).

 1
max_df int | float

Maximum document frequency (int or float).

 1.0
ngram_range tuple[int, int]

Tuple (min_n, max_n) for n-gram extraction.

 (1, 1)
Contract
  • vocabulary raises RuntimeError if accessed before fit.
  • Output matrix dtype is always float64.
Example

emb = CountVectorizerEmbedder() mat = emb.fit_transform(["the cat sat", "the dog ran"]) mat.shape[0] 2

Source code in kenon/embeddings.py 
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
class CountVectorizerEmbedder:
    """Corpus-internal count-based token embeddings via sklearn CountVectorizer.

    Args:
        stopwords: Stopword set to exclude. Pass ``None`` to keep all tokens.
        min_df: Minimum document frequency (int or float).
        max_df: Maximum document frequency (int or float).
        ngram_range: Tuple ``(min_n, max_n)`` for n-gram extraction.

    Contract:
        - ``vocabulary`` raises ``RuntimeError`` if accessed before ``fit``.
        - Output matrix dtype is always float64.

    Example:
        >>> emb = CountVectorizerEmbedder()
        >>> mat = emb.fit_transform(["the cat sat", "the dog ran"])
        >>> mat.shape[0]
        2
    """

    def __init__(
        self,
        stopwords: frozenset[str] | None = None,
        min_df: int | float = 1,
        max_df: int | float = 1.0,
        ngram_range: tuple[int, int] = (1, 1),
    ) -> None:
        stop = list(stopwords) if stopwords else None
        self._vectorizer = CountVectorizer(
            stop_words=stop,
            min_df=min_df,
            max_df=max_df,
            ngram_range=ngram_range,
        )
        self._fitted = False

    def fit(self, corpus: list[str]) -> None:
        """Fit the embedder on a corpus of strings.

        Args:
            corpus: List of document strings.
        """
        self._vectorizer.fit(corpus)
        self._fitted = True

    def transform(self, corpus: list[str]) -> Matrix:
        """Return embedding matrix (n_docs x n_features).

        Args:
            corpus: List of document strings.

        Returns:
            A 2-D float64 numpy array.
        """
        return self._vectorizer.transform(corpus).toarray().astype(np.float64)

    def fit_transform(self, corpus: list[str]) -> Matrix:
        """Fit and transform in one step.

        Args:
            corpus: List of document strings.

        Returns:
            A 2-D float64 numpy array.
        """
        result = self._vectorizer.fit_transform(corpus).toarray().astype(np.float64)
        self._fitted = True
        return result

    @property
    def vocabulary(self) -> dict[str, int]:
        """Mapping from token to column index.

        Raises:
            RuntimeError: If accessed before ``fit``.
        """
        if not self._fitted:
            msg = "Embedder has not been fitted yet. Call fit() first."
            raise RuntimeError(msg)
        return dict(self._vectorizer.vocabulary_)

vocabulary property

Mapping from token to column index.

Raises:

Type Description
RuntimeError

If accessed before fit.

fit(corpus)

Fit the embedder on a corpus of strings.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required
Source code in kenon/embeddings.py 
80
81
82
83
84
85
86
87
def fit(self, corpus: list[str]) -> None:
    """Fit the embedder on a corpus of strings.

    Args:
        corpus: List of document strings.
    """
    self._vectorizer.fit(corpus)
    self._fitted = True

fit_transform(corpus)

Fit and transform in one step.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array.
Source code in kenon/embeddings.py 
100
101
102
103
104
105
106
107
108
109
110
111
def fit_transform(self, corpus: list[str]) -> Matrix:
    """Fit and transform in one step.

    Args:
        corpus: List of document strings.

    Returns:
        A 2-D float64 numpy array.
    """
    result = self._vectorizer.fit_transform(corpus).toarray().astype(np.float64)
    self._fitted = True
    return result

transform(corpus)

Return embedding matrix (n_docs x n_features).

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array.
Source code in kenon/embeddings.py 
89
90
91
92
93
94
95
96
97
98
def transform(self, corpus: list[str]) -> Matrix:
    """Return embedding matrix (n_docs x n_features).

    Args:
        corpus: List of document strings.

    Returns:
        A 2-D float64 numpy array.
    """
    return self._vectorizer.transform(corpus).toarray().astype(np.float64)

kenon.embeddings.TfidfEmbedder

Corpus-internal TF-IDF token embeddings via sklearn TfidfVectorizer.

Parameters:

Name Type Description Default
stopwords frozenset[str] | None

Stopword set to exclude.

 None
min_df int | float

Minimum document frequency.

 1
max_df int | float

Maximum document frequency.

 1.0
sublinear_tf bool

Apply sublinear TF scaling (1 + log(tf)).

 False
Contract
  • vocabulary raises RuntimeError if accessed before fit.
  • Output matrix dtype is always float64.
Example

emb = TfidfEmbedder() mat = emb.fit_transform(["the cat sat", "the dog ran"]) mat.shape[0] 2

Source code in kenon/embeddings.py 
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
class TfidfEmbedder:
    """Corpus-internal TF-IDF token embeddings via sklearn TfidfVectorizer.

    Args:
        stopwords: Stopword set to exclude.
        min_df: Minimum document frequency.
        max_df: Maximum document frequency.
        sublinear_tf: Apply sublinear TF scaling (``1 + log(tf)``).

    Contract:
        - ``vocabulary`` raises ``RuntimeError`` if accessed before ``fit``.
        - Output matrix dtype is always float64.

    Example:
        >>> emb = TfidfEmbedder()
        >>> mat = emb.fit_transform(["the cat sat", "the dog ran"])
        >>> mat.shape[0]
        2
    """

    def __init__(
        self,
        stopwords: frozenset[str] | None = None,
        min_df: int | float = 1,
        max_df: int | float = 1.0,
        sublinear_tf: bool = False,
    ) -> None:
        stop = list(stopwords) if stopwords else None
        self._vectorizer = TfidfVectorizer(
            stop_words=stop,
            min_df=min_df,
            max_df=max_df,
            sublinear_tf=sublinear_tf,
        )
        self._fitted = False

    def fit(self, corpus: list[str]) -> None:
        """Fit the embedder on a corpus of strings.

        Args:
            corpus: List of document strings.
        """
        self._vectorizer.fit(corpus)
        self._fitted = True

    def transform(self, corpus: list[str]) -> Matrix:
        """Return embedding matrix (n_docs x n_features).

        Args:
            corpus: List of document strings.

        Returns:
            A 2-D float64 numpy array.
        """
        return self._vectorizer.transform(corpus).toarray().astype(np.float64)

    def fit_transform(self, corpus: list[str]) -> Matrix:
        """Fit and transform in one step.

        Args:
            corpus: List of document strings.

        Returns:
            A 2-D float64 numpy array.
        """
        result = self._vectorizer.fit_transform(corpus).toarray().astype(np.float64)
        self._fitted = True
        return result

    @property
    def vocabulary(self) -> dict[str, int]:
        """Mapping from token to column index.

        Raises:
            RuntimeError: If accessed before ``fit``.
        """
        if not self._fitted:
            msg = "Embedder has not been fitted yet. Call fit() first."
            raise RuntimeError(msg)
        return dict(self._vectorizer.vocabulary_)

vocabulary property

Mapping from token to column index.

Raises:

Type Description
RuntimeError

If accessed before fit.

fit(corpus)

Fit the embedder on a corpus of strings.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required
Source code in kenon/embeddings.py 
162
163
164
165
166
167
168
169
def fit(self, corpus: list[str]) -> None:
    """Fit the embedder on a corpus of strings.

    Args:
        corpus: List of document strings.
    """
    self._vectorizer.fit(corpus)
    self._fitted = True

fit_transform(corpus)

Fit and transform in one step.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array.
Source code in kenon/embeddings.py 
182
183
184
185
186
187
188
189
190
191
192
193
def fit_transform(self, corpus: list[str]) -> Matrix:
    """Fit and transform in one step.

    Args:
        corpus: List of document strings.

    Returns:
        A 2-D float64 numpy array.
    """
    result = self._vectorizer.fit_transform(corpus).toarray().astype(np.float64)
    self._fitted = True
    return result

transform(corpus)

Return embedding matrix (n_docs x n_features).

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array.
Source code in kenon/embeddings.py 
171
172
173
174
175
176
177
178
179
180
def transform(self, corpus: list[str]) -> Matrix:
    """Return embedding matrix (n_docs x n_features).

    Args:
        corpus: List of document strings.

    Returns:
        A 2-D float64 numpy array.
    """
    return self._vectorizer.transform(corpus).toarray().astype(np.float64)

kenon.embeddings.PMIEmbedder

Corpus-internal PPMI embeddings via chronowords.algebra.SVDAlgebra.

Builds a Positive Pointwise Mutual Information matrix from the supplied corpus using chronowords' Cython-optimised kernel, then applies SVD to produce dense word vectors. All statistics are derived exclusively from the supplied text — no external training corpus is involved.

Note

SVDAlgebra.train() accepts a generator of text lines and tokenises internally by whitespace splitting. It filters words shorter than min_word_length (default 3). The transform() method on this class returns word-level embeddings (vocab x n_components), not document-level embeddings.

Parameters:

Name Type Description Default
n_components int

Number of SVD dimensions (i.e. embedding size).

 100
window int

Context window size in tokens passed to SVDAlgebra.

 5
min_word_length int

Minimum word length for vocabulary inclusion.

 3

Raises:

Type Description
ImportError

If chronowords is not installed. Install with: uv add chronowords
Contract
  • vocabulary raises RuntimeError if accessed before fit.
  • transform() returns a 2-D float64 array of shape (len(vocabulary), n_components).
  • The embedder is serialisable via pickle.
Example

emb = PMIEmbedder(n_components=50, window=3) corpus = ["the cat sat on the mat", "the dog ran on the road"] mat = emb.fit_transform(corpus) mat.ndim 2

Source code in kenon/embeddings.py 
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
class PMIEmbedder:
    """Corpus-internal PPMI embeddings via ``chronowords.algebra.SVDAlgebra``.

    Builds a Positive Pointwise Mutual Information matrix from the supplied
    corpus using chronowords' Cython-optimised kernel, then applies SVD to
    produce dense word vectors. All statistics are derived exclusively from
    the supplied text — no external training corpus is involved.

    Note:
        ``SVDAlgebra.train()`` accepts a generator of text lines and tokenises
        internally by whitespace splitting. It filters words shorter than
        ``min_word_length`` (default 3). The ``transform()`` method on this
        class returns word-level embeddings (vocab x n_components), not
        document-level embeddings.

    Args:
        n_components: Number of SVD dimensions (i.e. embedding size).
        window: Context window size in tokens passed to ``SVDAlgebra``.
        min_word_length: Minimum word length for vocabulary inclusion.

    Raises:
        ImportError: If ``chronowords`` is not installed.
            Install with: ``uv add chronowords``

    Contract:
        - ``vocabulary`` raises ``RuntimeError`` if accessed before ``fit``.
        - ``transform()`` returns a 2-D float64 array of shape
          ``(len(vocabulary), n_components)``.
        - The embedder is serialisable via ``pickle``.

    Example:
        >>> emb = PMIEmbedder(n_components=50, window=3)
        >>> corpus = ["the cat sat on the mat", "the dog ran on the road"]
        >>> mat = emb.fit_transform(corpus)
        >>> mat.ndim
        2
    """

    def __init__(
        self,
        n_components: int = 100,
        window: int = 5,
        min_word_length: int = 3,
    ) -> None:
        try:
            from chronowords.algebra.svd import SVDAlgebra
        except ImportError as exc:
            msg = (
                "chronowords is required for PMIEmbedder. "
                "Install with: uv add chronowords"
            )
            raise ImportError(msg) from exc

        self._model = SVDAlgebra(
            n_components=n_components,
            window_size=window,
            min_word_length=min_word_length,
        )
        self._fitted = False

    def fit(self, corpus: list[str]) -> None:
        """Fit the embedder on a corpus of strings.

        Args:
            corpus: List of document strings. Each string is treated as a
                line of text; chronowords tokenises by whitespace internally.
        """
        self._model.train(line for line in corpus)
        self._fitted = True

    def transform(self, corpus: list[str]) -> Matrix:
        """Return word embedding matrix (n_vocab x n_components).

        Unlike the sklearn-based embedders which return document-level
        matrices, this returns word-level embeddings since PMI operates
        at the word level.

        Args:
            corpus: Ignored if already fitted. Provided for protocol
                compatibility.

        Returns:
            A 2-D float64 numpy array of shape ``(len(vocabulary), n_components)``.

        Raises:
            RuntimeError: If the embedder has not been fitted.
        """
        if not self._fitted or self._model.embeddings is None:
            msg = "Embedder has not been fitted yet. Call fit() first."
            raise RuntimeError(msg)
        return self._model.embeddings.astype(np.float64)

    def fit_transform(self, corpus: list[str]) -> Matrix:
        """Fit and transform in one step.

        Args:
            corpus: List of document strings.

        Returns:
            A 2-D float64 numpy array.
        """
        self.fit(corpus)
        return self.transform(corpus)

    @property
    def vocabulary(self) -> dict[str, int]:
        """Mapping from token to column index.

        Raises:
            RuntimeError: If accessed before ``fit``.
        """
        if not self._fitted:
            msg = "Embedder has not been fitted yet. Call fit() first."
            raise RuntimeError(msg)
        return dict(self._model._vocab_index)

vocabulary property

Mapping from token to column index.

Raises:

Type Description
RuntimeError

If accessed before fit.

fit(corpus)

Fit the embedder on a corpus of strings.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings. Each string is treated as a line of text; chronowords tokenises by whitespace internally.

 required
Source code in kenon/embeddings.py 
268
269
270
271
272
273
274
275
276
def fit(self, corpus: list[str]) -> None:
    """Fit the embedder on a corpus of strings.

    Args:
        corpus: List of document strings. Each string is treated as a
            line of text; chronowords tokenises by whitespace internally.
    """
    self._model.train(line for line in corpus)
    self._fitted = True

fit_transform(corpus)

Fit and transform in one step.

Parameters:

Name Type Description Default
corpus list[str]

List of document strings.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array.
Source code in kenon/embeddings.py 
300
301
302
303
304
305
306
307
308
309
310
def fit_transform(self, corpus: list[str]) -> Matrix:
    """Fit and transform in one step.

    Args:
        corpus: List of document strings.

    Returns:
        A 2-D float64 numpy array.
    """
    self.fit(corpus)
    return self.transform(corpus)

transform(corpus)

Return word embedding matrix (n_vocab x n_components).

Unlike the sklearn-based embedders which return document-level matrices, this returns word-level embeddings since PMI operates at the word level.

Parameters:

Name Type Description Default
corpus list[str]

Ignored if already fitted. Provided for protocol compatibility.

 required

Returns:

Type Description
Matrix

A 2-D float64 numpy array of shape (len(vocabulary), n_components).

Raises:

Type Description
RuntimeError

If the embedder has not been fitted.
Source code in kenon/embeddings.py 
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
def transform(self, corpus: list[str]) -> Matrix:
    """Return word embedding matrix (n_vocab x n_components).

    Unlike the sklearn-based embedders which return document-level
    matrices, this returns word-level embeddings since PMI operates
    at the word level.

    Args:
        corpus: Ignored if already fitted. Provided for protocol
            compatibility.

    Returns:
        A 2-D float64 numpy array of shape ``(len(vocabulary), n_components)``.

    Raises:
        RuntimeError: If the embedder has not been fitted.
    """
    if not self._fitted or self._model.embeddings is None:
        msg = "Embedder has not been fitted yet. Call fit() first."
        raise RuntimeError(msg)
    return self._model.embeddings.astype(np.float64)