Skip to content

Pandas

symconstraints.pandas

Integration with Pandas to aid in data cleaning dataframes.

symbols(df, symbol_list, **kwargs)

Return SymPy symbols with assumptions inferred from the dataframe dtypes.

Currently, it infers the following:

  • Unsigned integer dtypes are inferred to be nonnegative integers
  • Float dtypes are inferred to be real numbers
  • Complex dtypes are inferred to be complex numbers

Parameters:

Name Type Description Default
df DataFrame

Dataframe to infer symbol assumptions from.

 required
symbol_list str | list[str]

List of columns, can be represented as a space separated string.

 required
**kwargs

Extra arguments to be passed to sympy.Symbol

 {}

Returns:

Type Description
Symbol | list[Symbol]

SymPy symbols with assumptions inferred from the dataframe dtypes, corresponding to each column given. Returns a list if multiple columns are given, or a single symbol if a signle column is given.

Raises:

Type Description
ValueError

Raises a ValueError if a column is not found or an unsupported dtype is found.

Examples:

>>> import pandas as pd
>>> from symconstraints.pandas import symbols
>>> df = pd.DataFrame({
...    'Level': [1],
...    'Width': [5.3],
...    'Height': [7.6],
...    'Voltage': [5+3j]
... }).astype({'Level': 'uint8'})
>>> level, width, voltage = symbols(df, 'Level Width Voltage')
>>> level.is_nonnegative, level.is_integer
(True, True)
>>> width.is_integer, width.is_real
(None, True)
>>> voltage.is_real, voltage.is_complex
(None, True)
Source code in src/symconstraints/pandas.py 
 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
 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
def symbols(
    df: pandas.DataFrame, symbol_list: str | list[str], **kwargs
) -> Symbol | list[Symbol]:
    """Return SymPy symbols with assumptions inferred from the dataframe dtypes.

    Currently, it infers the following:

    * Unsigned integer dtypes are inferred to be nonnegative integers
    * Float dtypes are inferred to be real numbers
    * Complex dtypes are inferred to be complex numbers

    Parameters
    ----------
    df : pandas.DataFrame
        Dataframe to infer symbol assumptions from.
    symbol_list : str | list[str]
        List of columns, can be represented as a space separated string.
    **kwargs: dict, optional
        Extra arguments to be passed to `sympy.Symbol`

    Returns
    -------
    Symbol | list[Symbol]
        SymPy symbols with assumptions inferred from the dataframe dtypes, corresponding to each column given.
        Returns a list if multiple columns are given, or a single symbol if a signle column is given.

    Raises
    ------
    ValueError
        Raises a ValueError if a column is not found or an unsupported dtype is found.

    Examples
    --------
    >>> import pandas as pd
    >>> from symconstraints.pandas import symbols
    >>> df = pd.DataFrame({
    ...    'Level': [1],
    ...    'Width': [5.3],
    ...    'Height': [7.6],
    ...    'Voltage': [5+3j]
    ... }).astype({'Level': 'uint8'})
    >>> level, width, voltage = symbols(df, 'Level Width Voltage')
    >>> level.is_nonnegative, level.is_integer
    (True, True)
    >>> width.is_integer, width.is_real
    (None, True)
    >>> voltage.is_real, voltage.is_complex
    (None, True)
    """
    result = []
    symbols = symbol_list if isinstance(symbol_list, list) else symbol_list.split()
    for symbol in symbols:
        symbol_dtype = df.dtypes.get(symbol)
        if symbol_dtype is None:
            raise ValueError(
                f"Column {symbol} does not exist in the dataframe provided. Available columns are {list(df.columns)}."
            )

        assumptions = (
            {"integer": True, "nonnegative": True}
            if is_unsigned_integer_dtype(symbol_dtype)
            else {"integer": True}
            if is_integer_dtype(symbol_dtype)
            else {"real": True}
            if is_float_dtype(symbol_dtype)
            else {"complex": True}
            if is_complex_dtype(symbol_dtype)
            else None
        )

        if assumptions is None:
            raise ValueError(
                f"Unsupported data type {symbol_dtype} in column {symbol}."
            )

        result.append(Symbol(symbol, **assumptions, **kwargs))

    match len(result):
        case 0:
            raise ValueError("No symbols given.")
        case 1:
            return result[0]

    return result

check(constraints, df)

Return a table checking all the validations provided.

Parameters:

Name Type Description Default
constraints Constraints | Validation

Constraints or Validation to use for checking.

 required
df DataFrame

Dataframe to check

 required

Returns:

Type Description
DataFrame

Returns a dataframe showing the result of all the validations for each row in the dataframe. Each column in the dataframe corresponds to each validation operation provided. Each row corresponds to each row in the original dataframe. A result of 1 is shown for successful validation, 0 for an unsuccessful validation, and NaN for validations that can't be computed due to missing values. The result dtype is float16.

If a Constraints object is provided, the columns are pandas.MultiIndex, where the top level is a set of columns with all validations concerning those sets of columns under it.

Examples:

>>> import pandas as pd
>>> from symconstraints.pandas import symbols, check
>>> from symconstraints import Constraints
>>> from sympy import Eq
>>> df = pd.DataFrame({
...     'A': [5,6,8,9],
...     'B': [3,5,90,None],
...     'C': [14, 30, None, None]
... }, dtype=float)
>>> A, B, C = symbols(df, ['A', 'B', 'C'])
>>> constraints = Constraints([A > B, Eq(C, B*A)])
>>> check(constraints, df)
# Order may differ
  (A, B)  (C, A, B)  (C, B)  (C, A)
   A > B Eq(C, A*B) B < C/B A > C/A
0    1.0        0.0     1.0     1.0
1    1.0        1.0     1.0     1.0
2    0.0        NaN     NaN     NaN
3    NaN        NaN     NaN     NaN
Source code in src/symconstraints/pandas.py 
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
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
206
207
208
209
210
211
212
213
214
def check(
    constraints: Constraints | Validation, df: pandas.DataFrame
) -> pandas.DataFrame:
    """Return a table checking all the validations provided.

    Parameters
    ----------
    constraints : Constraints | Validation
        `Constraints` or `Validation` to use for checking.
    df : pandas.DataFrame
        Dataframe to check

    Returns
    -------
    pandas.DataFrame
        Returns a dataframe showing the result of all the validations for each row in the dataframe.
        Each column in the dataframe corresponds to each validation operation provided. Each row
        corresponds to each row in the original dataframe. A result of 1 is shown for successful validation,
        0 for an unsuccessful validation, and NaN for validations that can't be computed due to missing values.
        The result dtype is `float16`.

        If a `Constraints` object is provided, the columns are `pandas.MultiIndex`, where the top level is a set of
        columns with all validations concerning those sets of columns under it.

    Examples
    --------
    >>> import pandas as pd
    >>> from symconstraints.pandas import symbols, check
    >>> from symconstraints import Constraints
    >>> from sympy import Eq
    >>> df = pd.DataFrame({
    ...     'A': [5,6,8,9],
    ...     'B': [3,5,90,None],
    ...     'C': [14, 30, None, None]
    ... }, dtype=float)
    >>> A, B, C = symbols(df, ['A', 'B', 'C'])
    >>> constraints = Constraints([A > B, Eq(C, B*A)])
    >>> check(constraints, df)
    # Order may differ
      (A, B)  (C, A, B)  (C, B)  (C, A)
       A > B Eq(C, A*B) B < C/B A > C/A
    0    1.0        0.0     1.0     1.0
    1    1.0        1.0     1.0     1.0
    2    0.0        NaN     NaN     NaN
    3    NaN        NaN     NaN     NaN
    """
    if isinstance(constraints, Validation):
        columns = tuple(constraints.keys)
        index: list[Boolean] = list(constraints.operations)
        columns_str = [str(column) for column in columns]

        relevant_indices = ~df[columns_str].isna().any(axis="columns")
        if not isinstance(relevant_indices, pandas.Series):
            raise RuntimeError(
                f"unexpected type for relevant_indices: {type(relevant_indices)}"
            )

        relevant_df = df[relevant_indices]
        # Return a dataframe if the output is a series
        return pandas.DataFrame(
            pandas.concat(
                [
                    pandas.concat(
                        dict(
                            (
                                operation,
                                _lambdify(columns, operation)(
                                    *(relevant_df[str(column)] for column in columns)
                                ).astype("float16"),
                            )
                            for operation in index
                        ),
                        axis="columns",
                    ),
                    pandas.concat(
                        dict(
                            (
                                operation,
                                pandas.Series(
                                    math.nan,
                                    index=relevant_indices.index[~relevant_indices],
                                    copy=True,
                                ),
                            )
                            for operation in index
                        ),
                        axis="columns",
                    ),
                ],
                axis="rows",
            )
        )

    if isinstance(constraints, Constraints):
        return pandas.concat(
            dict(
                (validation.keys, check(validation, df))
                for validation in constraints.validations
            ),
            axis="columns",
        )

    raise ValueError(f"Invalid constraints given: {constraints}")

set_invalid_all(check_result, df, fill=math.nan)

Replace all possible invalid values in the dataframe to a set value.

This replaces values in the dataframe that could possibly be invalid under the given constraints. This might help get rid of outlier data within the dataframe.

The input dataframe is copied and is not edited in-place.

Parameters:

Name Type Description Default
check_result DataFrame

Check result returned by check_result

 required
df DataFrame

Dataframe to edit

 required
fill Any

The set value to replace invalid values.

 nan

Returns:

Type Description
DataFrame

Dataframe with replaced values.

Examples:

>>> import pandas as pd
>>> from symconstraints import Constraints
>>> from symconstraints.pandas import symbols, check, set_invalid_all
>>> from sympy import Eq
>>> df = pd.DataFrame(
...    {
...         "height": [5, 6, 8, 9],
...         "width": [3, 5, 90, None],
...         "area": [14, 30, None, 18],
...     },
...     dtype=float,
... )
>>> height, width, area = symbols(df, ["height", "width", "area"])
>>> constraints = Constraints([height > width, Eq(area, width * height)])
>>> check_result = check(constraints, df)
>>> set_invalid_all(check_result, df)
   height  width  area
0     NaN    NaN   NaN
1     6.0    5.0  30.0
2     NaN    NaN   NaN
3     9.0    NaN  18.0
Source code in src/symconstraints/pandas.py 
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
def set_invalid_all(
    check_result: pandas.DataFrame, df: pandas.DataFrame, fill=math.nan
) -> pandas.DataFrame:
    """Replace all possible invalid values in the dataframe to a set value.

    This replaces values in the dataframe that could possibly be invalid under the given constraints. This might help
    get rid of outlier data within the dataframe.

    The input dataframe is copied and is not edited in-place.

    Parameters
    ----------
    check_result : pandas.DataFrame
        Check result returned by `check_result`
    df : pandas.DataFrame
        Dataframe to edit
    fill : Any, optional
        The set value to replace invalid values.

    Returns
    -------
    pandas.DataFrame
        Dataframe with replaced values.

    Examples
    --------
    >>> import pandas as pd
    >>> from symconstraints import Constraints
    >>> from symconstraints.pandas import symbols, check, set_invalid_all
    >>> from sympy import Eq
    >>> df = pd.DataFrame(
    ...    {
    ...         "height": [5, 6, 8, 9],
    ...         "width": [3, 5, 90, None],
    ...         "area": [14, 30, None, 18],
    ...     },
    ...     dtype=float,
    ... )
    >>> height, width, area = symbols(df, ["height", "width", "area"])
    >>> constraints = Constraints([height > width, Eq(area, width * height)])
    >>> check_result = check(constraints, df)
    >>> set_invalid_all(check_result, df)
       height  width  area
    0     NaN    NaN   NaN
    1     6.0    5.0  30.0
    2     NaN    NaN   NaN
    3     9.0    NaN  18.0
    """
    result = df.copy()
    if check_result.empty:
        return result
    if check_result.columns.nlevels == 1:
        column_item = check_result.columns.item()
        if not isinstance(column_item, Boolean):
            raise ValueError(
                f"Invalid check result given. It has a column {column_item} of type: {type(column_item)}"
            )
        result.loc[
            ~check_result.fillna(1.0).all(axis="columns"),
            [str(symbol) for symbol in column_item.free_symbols],
        ] = fill

        return result

    keysets = check_result.columns.get_level_values(0)

    invalid_keysets = [keys for keys in keysets if not isinstance(keys, frozenset)]
    if len(invalid_keysets) > 0:
        raise ValueError(f"Found invalid columns in check result: {invalid_keysets}")

    for keys in keysets:
        result.loc[
            ~check_result[keys].fillna(1.0).all(axis="columns"),
            [str(key) for key in keys],
        ] = fill

    return result

set_invalid_min(check_result, df, fill=math.nan, priority=None)

Replace the minimum amount of possible invalid values in the dataframe to a set value.

Similar to set_invalid_all, this replaces values in the dataframe that could possibly be invalid under the given constraints. However it does not replace all the values, instead it tries to replace the minimum amount of values in each row such that it satisfies all the constraints. This might help get rid of outlier data within the dataframe, while also keeping more of the values in the dataset unchanged.

The input dataframe is copied and is not edited in-place.

Parameters:

Name Type Description Default
check_result DataFrame

Check result returned by check_result

 required
df DataFrame

Dataframe to edit

 required
fill Any

The set value to replace invalid values.

 nan
priority list[Symbol] | None

In case of a tie in the decision to set which column, refer to this priority to decide which column to to set first. Priority is ordered from most preferred to be set to least preferred.

 None

Returns:

Type Description
DataFrame

Dataframe with replaced values.

Examples:

>>> import pandas as pd
>>> from symconstraints import Constraints
>>> from symconstraints.pandas import symbols, check, set_invalid_min
>>> from sympy import Eq
>>> df = pd.DataFrame(
...    {
...        "height": [5, 6, 8, 9, 7],
...        "width": [3, 5, 90, None, 8],
...        "depth": [None, 2, 10, 5, 5],
...        "area": [14, 30, 10, None, 35],
...        "volume": [None, 60, 100, 30, None],
...    },
...    dtype=float,
... )
>>> height, width, area, depth, volume = symbols(df, ["height", "width", "area", "depth", "volume"])
>>> constraints = Constraints([
...     Eq(area, height * width),
...     Eq(volume, area * depth),
...     height > width,
...     width > depth,
... ])
>>> check_result = check(constraints, df)
>>> set_invalid_min(check_result, df, priority=[volume, area, depth, width, height])
   height  width  depth  area  volume
0     5.0    3.0    NaN   NaN     NaN
1     6.0    5.0    2.0  30.0    60.0
2     NaN   90.0   10.0   NaN   100.0
3     9.0    NaN    5.0   NaN    30.0
4     7.0    NaN    5.0   NaN     NaN
Source code in src/symconstraints/pandas.py 
309
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
363
364
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
424
425
426
427
428
429
430
431
432
433
def set_invalid_min(
    check_result: pandas.DataFrame,
    df: pandas.DataFrame,
    fill=math.nan,
    priority: list[Symbol] | None = None,
) -> pandas.DataFrame:
    """Replace the minimum amount of possible invalid values in the dataframe to a set value.

    Similar to `set_invalid_all`, this replaces values in the dataframe that could possibly be invalid under the given constraints.
    However it does not replace all the values, instead it tries to replace the minimum amount of values in each row such that
    it satisfies all the constraints. This might help get rid of outlier data within the dataframe, while also keeping more of the values
    in the dataset unchanged.

    The input dataframe is copied and is not edited in-place.

    Parameters
    ----------
    check_result : pandas.DataFrame
        Check result returned by `check_result`
    df : pandas.DataFrame
        Dataframe to edit
    fill : Any, optional
        The set value to replace invalid values.
    priority : list[Symbol] | None, optional
        In case of a tie in the decision to set which column, refer to this priority to decide which
        column to to set first. Priority is ordered from most preferred to be set to least preferred.

    Returns
    -------
    pandas.DataFrame
        Dataframe with replaced values.

    Examples
    --------
    >>> import pandas as pd
    >>> from symconstraints import Constraints
    >>> from symconstraints.pandas import symbols, check, set_invalid_min
    >>> from sympy import Eq
    >>> df = pd.DataFrame(
    ...    {
    ...        "height": [5, 6, 8, 9, 7],
    ...        "width": [3, 5, 90, None, 8],
    ...        "depth": [None, 2, 10, 5, 5],
    ...        "area": [14, 30, 10, None, 35],
    ...        "volume": [None, 60, 100, 30, None],
    ...    },
    ...    dtype=float,
    ... )
    >>> height, width, area, depth, volume = symbols(df, ["height", "width", "area", "depth", "volume"])
    >>> constraints = Constraints([
    ...     Eq(area, height * width),
    ...     Eq(volume, area * depth),
    ...     height > width,
    ...     width > depth,
    ... ])
    >>> check_result = check(constraints, df)
    >>> set_invalid_min(check_result, df, priority=[volume, area, depth, width, height])
       height  width  depth  area  volume
    0     5.0    3.0    NaN   NaN     NaN
    1     6.0    5.0    2.0  30.0    60.0
    2     NaN   90.0   10.0   NaN   100.0
    3     9.0    NaN    5.0   NaN    30.0
    4     7.0    NaN    5.0   NaN     NaN
    """
    validation_sets = check_result.columns.get_level_values(0).unique()
    symbols = reduce(or_, validation_sets, frozenset())
    symbols_str = [str(s) for s in symbols]

    if priority is not None:
        priority_indices = dict(
            (str(value), index) for index, value in enumerate(priority)
        )
        symbols_str.sort(key=lambda symbol: priority_indices.get(symbol, math.inf))

    # Finding the minimum number of symbols to replace is equivalent to the set cover problem.
    check_sets = check_result.groupby(level=0, axis="columns").all()
    check_sets.columns = check_sets.columns.get_level_values(0)

    set_to_symbols = pandas.DataFrame(
        [
            dict(
                (symbol, symbol in frozenset(str(s) for s in subset))
                for symbol in symbols_str
            )
            for subset in validation_sets
        ],
        index=validation_sets,
    )

    set_cover_results = pandas.DataFrame(columns=[*validation_sets, *symbols_str])

    for _, selection in check_sets.drop_duplicates().iterrows():
        invalid_sets = selection.index[selection == 0.0]
        if not isinstance(invalid_sets, pandas.Index):
            raise ValueError(
                f"Unexpected value found when fetching invalid sets index: {invalid_sets}"
            )
        invalid_set_to_symbols = set_to_symbols.loc[invalid_sets, :]
        invalid_set_to_symbols = invalid_set_to_symbols[
            invalid_set_to_symbols.any(axis="columns")
        ]
        invalid_symbols = list(invalid_set_to_symbols.columns)
        invalid_set_to_symbols_numpy = invalid_set_to_symbols.to_numpy()
        symbols_to_remove = set(
            invalid_symbols[index]
            for index in _set_cover(
                invalid_set_to_symbols_numpy.tobytes(),
                invalid_set_to_symbols_numpy.shape,
            )
        )
        set_cover_results.loc[len(set_cover_results)] = pandas.concat(
            [
                selection,
                pandas.Series(
                    dict(
                        (symbol, symbol in symbols_to_remove) for symbol in symbols_str
                    )
                ),
            ]
        )

    to_set = check_sets.merge(set_cover_results, how="left")[symbols_str]
    to_set.index = check_sets.index  # I don't know why I have to do this.

    return df.where(~to_set, fill)

impute(constraints, df)

Impute the dataframe under the given constraints.

This returns a copy of the dataframe with all NA values replaced with values inferred from imputation operations given in the constraints. This assumes that all the values are valid, so it is recommended that the dataframe be checked and that all of its invalid values are removed via check and set_invalid_all.

The input dataframe is not edited in-place.

Parameters:

Name Type Description Default
constraints Constraints | Imputation

Constraints or Imputation to use for imputation.

 required
df DataFrame

Dataframe to impute.

 required

Returns:

Type Description
DataFrame

Imputed dataframe.

Examples:

>>> import pandas as pd
>>> from symconstraints import Constraints
>>> from symconstraints.pandas import symbols, check, set_invalid_all, impute
>>> from sympy import Eq
>>> df = pd.DataFrame(
...    {
...         "height": [5, 6, 8, 9],
...         "width": [3, 5, 7, None],
...         "area": [14, 30, None, 18],
...     },
...     dtype=float,
... )
>>> height, width, area = symbols(df, ["height", "width", "area"])
>>> constraints = Constraints([height > width, Eq(area, width * height)])
>>> check_result = check(constraints, df)
>>> df = set_invalid_all(check_result, df)
>>> df
   height  width  area
0     NaN    NaN   NaN
1     6.0    5.0  30.0
2     8.0    7.0   NaN
3     9.0    NaN  18.0
>>> imputed_df = impute(constraints, df)
>>> imputed_df
   height  width  area
0     NaN    NaN   NaN
1     6.0    5.0  30.0
2     8.0    7.0  56.0
3     9.0    2.0  18.0
Source code in src/symconstraints/pandas.py 
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
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
def impute(
    constraints: Constraints | Imputation, df: pandas.DataFrame
) -> pandas.DataFrame:
    """Impute the dataframe under the given constraints.

    This returns a copy of the dataframe with all NA values replaced with values inferred from
    imputation operations given in the constraints. This assumes that all the values are valid,
    so it is recommended that the dataframe be checked and that all of its invalid values are
    removed via `check` and `set_invalid_all`.

    The input dataframe is not edited in-place.

    Parameters
    ----------
    constraints : Constraints | Imputation
        `Constraints` or `Imputation` to use for imputation.
    df : pandas.DataFrame
        Dataframe to impute.

    Returns
    -------
    pandas.DataFrame
        Imputed dataframe.

    Examples
    --------
    >>> import pandas as pd
    >>> from symconstraints import Constraints
    >>> from symconstraints.pandas import symbols, check, set_invalid_all, impute
    >>> from sympy import Eq
    >>> df = pd.DataFrame(
    ...    {
    ...         "height": [5, 6, 8, 9],
    ...         "width": [3, 5, 7, None],
    ...         "area": [14, 30, None, 18],
    ...     },
    ...     dtype=float,
    ... )
    >>> height, width, area = symbols(df, ["height", "width", "area"])
    >>> constraints = Constraints([height > width, Eq(area, width * height)])
    >>> check_result = check(constraints, df)
    >>> df = set_invalid_all(check_result, df)
    >>> df
       height  width  area
    0     NaN    NaN   NaN
    1     6.0    5.0  30.0
    2     8.0    7.0   NaN
    3     9.0    NaN  18.0
    >>> imputed_df = impute(constraints, df)
    >>> imputed_df
       height  width  area
    0     NaN    NaN   NaN
    1     6.0    5.0  30.0
    2     8.0    7.0  56.0
    3     9.0    2.0  18.0
    """
    if isinstance(constraints, Imputation):
        result = df.copy()

        columns = tuple(constraints.keys)
        columns_str = [str(column) for column in columns]

        result.loc[
            result[columns_str].notna().all(axis="columns")
            & result[str(constraints.target_key)].isna(),
            str(constraints.target_key),
        ] = _lambdify(columns, constraints.operation)(
            *(result[column_str] for column_str in columns_str)
        )

        return result

    if isinstance(constraints, Constraints):
        result = df.copy()

        for imputation in constraints.imputations:
            columns = tuple(imputation.keys)
            columns_str = [str(column) for column in columns]

            result.loc[
                result[columns_str].notna().all(axis="columns")
                & result[str(imputation.target_key)].isna(),
                str(imputation.target_key),
            ] = _lambdify(columns, imputation.operation)(
                *(result[column_str] for column_str in columns_str)
            )

        return result

    raise ValueError(f"Invalid constraints given: {constraints}")