Skip to content

Similarity API

Basic Similarities

Raw Dot Product

Compute dot product similarity between rows of matrix1 and columns of matrix2.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix (e.g., user-item or item-user).

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
k int

Number of top-k items per row.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before similarity computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format: 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 means all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of shape (n_rows, n_cols) in the specified format,
spmatrix

containing the top-k dot product similarities.
Source code in similaripy/similarity.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
def dot_product(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute dot product similarity between rows of matrix1 and columns of matrix2.

    Args:
        matrix1: Input sparse matrix (e.g., user-item or item-user).
        matrix2: Optional second matrix. If None, uses matrix1.T.
        k: Number of top-k items per row.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before similarity computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format: 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 means all available cores).

    Returns:
        A sparse matrix of shape (n_rows, n_cols) in the specified format,
        containing the top-k dot product similarities.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
Cosine Similarity

Compute cosine similarity between sparse vectors.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k cosine similarities in the specified format.
Source code in similaripy/similarity.py 
 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
def cosine(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute cosine similarity between sparse vectors.

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k cosine similarities in the specified format.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l2=1,
        c1=0.5, c2=0.5,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
Asymmetric Cosine

Compute asymmetric cosine similarity.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix (e.g., user-item or item-user).

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
alpha float

Controls asymmetry in cosine weighting. alpha=1 weighs only matrix1; alpha=0.5 is symmetric.

 0.5
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of shape (n_rows, n_cols) containing the top-k
spmatrix

asymmetric cosine similarities in the specified format.
Source code in similaripy/similarity.py 
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
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
def asymmetric_cosine(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    alpha: float = 0.5,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute asymmetric cosine similarity.

    Args:
        matrix1: Input sparse matrix (e.g., user-item or item-user).
        matrix2: Optional second matrix. If None, uses matrix1.T.
        alpha: Controls asymmetry in cosine weighting.
               `alpha=1` weighs only matrix1; `alpha=0.5` is symmetric.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of shape (n_rows, n_cols) containing the top-k
        asymmetric cosine similarities in the specified format.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l2=1,
        c1=alpha, c2=1-alpha,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
Jaccard Similarity

Compute Jaccard similarity (intersection over union).

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k Jaccard similarities in the specified format.
Source code in similaripy/similarity.py 
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
def jaccard(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute Jaccard similarity (intersection over union).

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k Jaccard similarities in the specified format.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l1=1,
        t1=1, t2=1,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
Dice Similarity

Compute Dice similarity (harmonic mean of overlap and size).

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k Dice similarities in the specified format.
Source code in similaripy/similarity.py 
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
def dice(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute Dice similarity (harmonic mean of overlap and size).

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k Dice similarities in the specified format.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l1=1,
        t1=0.5, t2=0.5,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
Tversky Similarity

Compute Tversky similarity between sparse vectors.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
alpha float

Tversky weight for elements unique to matrix1.

 1.0
beta float

Tversky weight for elements unique to matrix2.

 1.0
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k Tversky similarities in the specified format.
Source code in similaripy/similarity.py 
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
206
207
208
209
def tversky(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    alpha: float = 1.0,
    beta: float = 1.0,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute Tversky similarity between sparse vectors.

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        alpha: Tversky weight for elements unique to matrix1.
        beta: Tversky weight for elements unique to matrix2.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k Tversky similarities in the specified format.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l1=1,
        t1=alpha, t2=beta,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )

Graph-Based Similarities

P3α

Compute P3alpha similarity using a normalized 3-step random walk.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
alpha float

Exponent for transition probabilities to control popularity effect.

 1.0
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k P3alpha similarities in the specified format.
Source code in similaripy/similarity.py 
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
def p3alpha(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    alpha: float = 1.0,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute P3alpha similarity using a normalized 3-step random walk.

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        alpha: Exponent for transition probabilities to control popularity effect.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k P3alpha similarities in the specified format.
    """
    if matrix2 is None:
        matrix2 = matrix1.T
    matrix1 = _normalize(matrix1, norm='l1', axis=1, inplace=False)
    matrix1.data = np.power(matrix1.data, alpha)
    matrix2 = _normalize(matrix2, norm='l1', axis=1, inplace=False)
    matrix2.data = np.power(matrix2.data, alpha)
    return _sim.s_plus(
        matrix1=matrix1, matrix2=matrix2,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )
RP3β

Compute RP3beta similarity: P3alpha with popularity penalization.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
alpha float

Exponent for transition probabilities.

 1.0
beta float

Exponent to penalize popularity based on column sums.

 1.0
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k RP3beta similarities in the specified format.
Source code in similaripy/similarity.py 
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
def rp3beta(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    alpha: float = 1.0,
    beta: float = 1.0,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute RP3beta similarity: P3alpha with popularity penalization.

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        alpha: Exponent for transition probabilities.
        beta: Exponent to penalize popularity based on column sums.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k RP3beta similarities in the specified format.
    """
    if matrix2 is None:
        matrix2 = matrix1.T
    pop_m2 = matrix2.sum(axis=0).A1
    matrix1 = _normalize(matrix1, norm='l1', axis=1, inplace=False)
    matrix1.data = np.power(matrix1.data, alpha)
    matrix2 = _normalize(matrix2, norm='l1', axis=1, inplace=False)
    matrix2.data = np.power(matrix2.data, alpha)
    return _sim.s_plus(
        matrix1=matrix1, matrix2=matrix2,
        weight_depop_matrix2=pop_m2,
        p2=beta,
        l3=1,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )

Hybrid Similarity

S Plus

Compute hybrid S Plus similarity with weighted Tversky and Cosine components.

Parameters:

Name Type Description Default
matrix1 spmatrix

Input sparse matrix.

 required
matrix2 Optional[spmatrix]

Optional second matrix. If None, uses matrix1.T.

 None
l float

Mixing parameter between Tversky (l1) and Cosine (l2).

 0.5
t1 float

Tversky alpha for matrix1.

 1.0
t2 float

Tversky beta for matrix2.

 1.0
c float

Cosine exponent coefficient.

 0.5
k int

Number of top-k items per row to keep.

 100
shrink float

Shrinkage value applied to similarity scores.

 0.0
threshold float

Minimum similarity value to retain.

 0.0
binary bool

Whether to binarize the input matrix before computation.

 False
target_rows Optional[Union[list[int], ndarray]]

List or array of row indices to compute. If None, computes all.

 None
target_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to include before top-k. Can be a list or sparse mask matrix.

 None
filter_cols Optional[Union[list[int], ndarray, spmatrix]]

Columns to exclude before top-k. Can be a list or sparse mask matrix.

 None
verbose bool

Whether to show a progress bar.

 True
format_output Literal['csr', 'coo']

Output format, either 'csr' or 'coo'. Use 'coo' on Windows.

 'coo'
num_threads int

Number of threads to use (0 = all available cores).

 0

Returns:

Type Description
spmatrix

A sparse matrix of top-k similarities based on combined Tversky and Cosine scoring.
Source code in similaripy/similarity.py 
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
def s_plus(
    matrix1: spmatrix,
    matrix2: Optional[spmatrix] = None,
    l: float = 0.5,
    t1: float = 1.0,
    t2: float = 1.0,
    c: float = 0.5,
    k: int = 100,
    shrink: float = 0.0,
    threshold: float = 0.0,
    binary: bool = False,
    target_rows: Optional[Union[list[int], np.ndarray]] = None,
    target_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    filter_cols: Optional[Union[list[int], np.ndarray, spmatrix]] = None,
    verbose: bool = True,
    format_output: Literal['csr', 'coo'] = 'coo',
    num_threads: int = 0
) -> spmatrix:
    """
    Compute hybrid S Plus similarity with weighted Tversky and Cosine components.

    Args:
        matrix1: Input sparse matrix.
        matrix2: Optional second matrix. If None, uses matrix1.T.
        l: Mixing parameter between Tversky (l1) and Cosine (l2).
        t1: Tversky alpha for matrix1.
        t2: Tversky beta for matrix2.
        c: Cosine exponent coefficient.
        k: Number of top-k items per row to keep.
        shrink: Shrinkage value applied to similarity scores.
        threshold: Minimum similarity value to retain.
        binary: Whether to binarize the input matrix before computation.
        target_rows: List or array of row indices to compute. If None, computes all.
        target_cols: Columns to include before top-k. Can be a list or sparse mask matrix.
        filter_cols: Columns to exclude before top-k. Can be a list or sparse mask matrix.
        verbose: Whether to show a progress bar.
        format_output: Output format, either 'csr' or 'coo'. Use 'coo' on Windows.
        num_threads: Number of threads to use (0 = all available cores).

    Returns:
        A sparse matrix of top-k similarities based on combined Tversky and Cosine scoring.
    """
    return _sim.s_plus(
        matrix1, matrix2=matrix2,
        l1=l, l2=1-l,
        t1=t1, t2=t2,
        c1=c, c2=1-c,
        k=k, shrink=shrink, threshold=threshold,
        binary=binary,
        target_rows=target_rows,
        target_cols=target_cols,
        filter_cols=filter_cols,
        verbose=verbose,
        format_output=format_output,
        num_threads=num_threads
    )