Skip to content

custom_filter_measurements

A polars-to-polars transformation function for filtering subjects by sequence length.

filter_measurements_fntr(stage_cfg, code_metadata, code_modifiers=None)

Returns a function that filters subject events to only encompass those with a set of permissible codes.

Parameters:

Name Type Description Default
df

The input DataFrame.

 required
stage_cfg DictConfig

The configuration for the code filtering stage.

 required

Returns:

Type Description
Callable[[LazyFrame], LazyFrame]

The processed DataFrame.

Examples:

>>> code_metadata_df = pl.DataFrame({
...     "code":               ["A", "A", "B", "C"],
...     "modifier1":          [1,   2,   1,   2],
...     "code/n_subjects":    [2,   1,   3,   2],
...     "code/n_occurrences": [4,   5,   3,   2],
... })
>>> data = pl.DataFrame({
...     "subject_id": [1,   1,   2,   2],
...     "code":       ["A", "B", "A", "C"],
...     "modifier1":  [1,   1,   2,   2],
... }).lazy()
>>> stage_cfg = DictConfig({"min_subjects_per_code": 2, "min_occurrences_per_code": 3})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (2, 3)
┌────────────┬──────┬───────────┐
│ subject_id ┆ code ┆ modifier1 │
│ ---        ┆ ---  ┆ ---       │
│ i64        ┆ str  ┆ i64       │
╞════════════╪══════╪═══════════╡
│ 1          ┆ A    ┆ 1         │
│ 1          ┆ B    ┆ 1         │
└────────────┴──────┴───────────┘
>>> stage_cfg = DictConfig({"min_subjects_per_code": 1, "min_occurrences_per_code": 4})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (2, 3)
┌────────────┬──────┬───────────┐
│ subject_id ┆ code ┆ modifier1 │
│ ---        ┆ ---  ┆ ---       │
│ i64        ┆ str  ┆ i64       │
╞════════════╪══════╪═══════════╡
│ 1          ┆ A    ┆ 1         │
│ 2          ┆ A    ┆ 2         │
└────────────┴──────┴───────────┘
>>> stage_cfg = DictConfig({"min_subjects_per_code": 1})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (4, 3)
┌────────────┬──────┬───────────┐
│ subject_id ┆ code ┆ modifier1 │
│ ---        ┆ ---  ┆ ---       │
│ i64        ┆ str  ┆ i64       │
╞════════════╪══════╪═══════════╡
│ 1          ┆ A    ┆ 1         │
│ 1          ┆ B    ┆ 1         │
│ 2          ┆ A    ┆ 2         │
│ 2          ┆ C    ┆ 2         │
└────────────┴──────┴───────────┘
>>> stage_cfg = DictConfig({"min_subjects_per_code": None, "min_occurrences_per_code": None})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (4, 3)
┌────────────┬──────┬───────────┐
│ subject_id ┆ code ┆ modifier1 │
│ ---        ┆ ---  ┆ ---       │
│ i64        ┆ str  ┆ i64       │
╞════════════╪══════╪═══════════╡
│ 1          ┆ A    ┆ 1         │
│ 1          ┆ B    ┆ 1         │
│ 2          ┆ A    ┆ 2         │
│ 2          ┆ C    ┆ 2         │
└────────────┴──────┴───────────┘
>>> stage_cfg = DictConfig({"min_occurrences_per_code": 5})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (1, 3)
┌────────────┬──────┬───────────┐
│ subject_id ┆ code ┆ modifier1 │
│ ---        ┆ ---  ┆ ---       │
│ i64        ┆ str  ┆ i64       │
╞════════════╪══════╪═══════════╡
│ 2          ┆ A    ┆ 2         │
└────────────┴──────┴───────────┘

>>> data = pl.DataFrame({
...     "subject_id": [1,   1,   2,   2],
...     "code":       ["BIRTH", "VISIT//PULMONARY", "BIRTH", "BIRTH//VISIT"],
...     "modifier1":  [1,   1,   2,   2],
... }).lazy()
>>> code_metadata_df = pl.DataFrame({
...     "code":               ["BIRTH", "VISIT//PULMONARY", "BIRTH", "BIRTH//VISIT"],
...     "modifier1":          [1,   1,   2,   2],
...     "code/n_subjects":    [2,   1,   3,   2],
...     "code/n_occurrences": [4,   5,   3,   2],
... })
>>> stage_cfg = DictConfig({"additional_codes": ["^BIRTH$", "^VISIT.*$"]})
>>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
>>> fn(data).collect()
shape: (3, 3)
┌────────────┬──────────────────┬───────────┐
│ subject_id ┆ code             ┆ modifier1 │
│ ---        ┆ ---              ┆ ---       │
│ i64        ┆ str              ┆ i64       │
╞════════════╪══════════════════╪═══════════╡
│ 1          ┆ BIRTH            ┆ 1         │
│ 1          ┆ VISIT//PULMONARY ┆ 1         │
│ 2          ┆ BIRTH            ┆ 2         │
└────────────┴──────────────────┴───────────┘
This stage works even if the default row index column exists

code_metadata_df = pl.DataFrame({ … “code”: [“A”, “A”, “B”, “C”], … “modifier1”: [1, 2, 1, 2], … “code/n_subjects”: [2, 1, 3, 2], … “code/n_occurrences”: [4, 5, 3, 2], … }) data = pl.DataFrame({ … “subject_id”: [1, 1, 2, 2], … “code”: [“A”, “B”, “A”, “C”], … “modifier1”: [1, 1, 2, 2], … “_row_idx”: [1, 1, 1, 1], … }).lazy() stage_cfg = DictConfig({“min_subjects_per_code”: 2, “min_occurrences_per_code”: 3}) fn = filter_measurements_fntr(stage_cfg, code_metadata_df, [“modifier1”]) fn(data).collect() shape: (2, 4) ┌────────────┬──────┬───────────┬──────────┐ │ subject_id ┆ code ┆ modifier1 ┆ _row_idx │ │ — ┆ — ┆ — ┆ — │ │ i64 ┆ str ┆ i64 ┆ i64 │ ╞════════════╪══════╪═══════════╪══════════╡ │ 1 ┆ A ┆ 1 ┆ 1 │ │ 1 ┆ B ┆ 1 ┆ 1 │ └────────────┴──────┴───────────┴──────────┘

Source code in meds_torch/utils/custom_filter_measurements.py 
 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
 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
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
def filter_measurements_fntr(
    stage_cfg: DictConfig, code_metadata: pl.LazyFrame, code_modifiers: list[str] | None = None
) -> Callable[[pl.LazyFrame], pl.LazyFrame]:
    """Returns a function that filters subject events to only encompass those with a set
    of permissible codes.

    Args:
        df: The input DataFrame.
        stage_cfg: The configuration for the code filtering stage.

    Returns:
        The processed DataFrame.

    Examples:
        >>> code_metadata_df = pl.DataFrame({
        ...     "code":               ["A", "A", "B", "C"],
        ...     "modifier1":          [1,   2,   1,   2],
        ...     "code/n_subjects":    [2,   1,   3,   2],
        ...     "code/n_occurrences": [4,   5,   3,   2],
        ... })
        >>> data = pl.DataFrame({
        ...     "subject_id": [1,   1,   2,   2],
        ...     "code":       ["A", "B", "A", "C"],
        ...     "modifier1":  [1,   1,   2,   2],
        ... }).lazy()
        >>> stage_cfg = DictConfig({"min_subjects_per_code": 2, "min_occurrences_per_code": 3})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (2, 3)
        ┌────────────┬──────┬───────────┐
        │ subject_id ┆ code ┆ modifier1 │
        │ ---        ┆ ---  ┆ ---       │
        │ i64        ┆ str  ┆ i64       │
        ╞════════════╪══════╪═══════════╡
        │ 1          ┆ A    ┆ 1         │
        │ 1          ┆ B    ┆ 1         │
        └────────────┴──────┴───────────┘
        >>> stage_cfg = DictConfig({"min_subjects_per_code": 1, "min_occurrences_per_code": 4})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (2, 3)
        ┌────────────┬──────┬───────────┐
        │ subject_id ┆ code ┆ modifier1 │
        │ ---        ┆ ---  ┆ ---       │
        │ i64        ┆ str  ┆ i64       │
        ╞════════════╪══════╪═══════════╡
        │ 1          ┆ A    ┆ 1         │
        │ 2          ┆ A    ┆ 2         │
        └────────────┴──────┴───────────┘
        >>> stage_cfg = DictConfig({"min_subjects_per_code": 1})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (4, 3)
        ┌────────────┬──────┬───────────┐
        │ subject_id ┆ code ┆ modifier1 │
        │ ---        ┆ ---  ┆ ---       │
        │ i64        ┆ str  ┆ i64       │
        ╞════════════╪══════╪═══════════╡
        │ 1          ┆ A    ┆ 1         │
        │ 1          ┆ B    ┆ 1         │
        │ 2          ┆ A    ┆ 2         │
        │ 2          ┆ C    ┆ 2         │
        └────────────┴──────┴───────────┘
        >>> stage_cfg = DictConfig({"min_subjects_per_code": None, "min_occurrences_per_code": None})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (4, 3)
        ┌────────────┬──────┬───────────┐
        │ subject_id ┆ code ┆ modifier1 │
        │ ---        ┆ ---  ┆ ---       │
        │ i64        ┆ str  ┆ i64       │
        ╞════════════╪══════╪═══════════╡
        │ 1          ┆ A    ┆ 1         │
        │ 1          ┆ B    ┆ 1         │
        │ 2          ┆ A    ┆ 2         │
        │ 2          ┆ C    ┆ 2         │
        └────────────┴──────┴───────────┘
        >>> stage_cfg = DictConfig({"min_occurrences_per_code": 5})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (1, 3)
        ┌────────────┬──────┬───────────┐
        │ subject_id ┆ code ┆ modifier1 │
        │ ---        ┆ ---  ┆ ---       │
        │ i64        ┆ str  ┆ i64       │
        ╞════════════╪══════╪═══════════╡
        │ 2          ┆ A    ┆ 2         │
        └────────────┴──────┴───────────┘

        >>> data = pl.DataFrame({
        ...     "subject_id": [1,   1,   2,   2],
        ...     "code":       ["BIRTH", "VISIT//PULMONARY", "BIRTH", "BIRTH//VISIT"],
        ...     "modifier1":  [1,   1,   2,   2],
        ... }).lazy()
        >>> code_metadata_df = pl.DataFrame({
        ...     "code":               ["BIRTH", "VISIT//PULMONARY", "BIRTH", "BIRTH//VISIT"],
        ...     "modifier1":          [1,   1,   2,   2],
        ...     "code/n_subjects":    [2,   1,   3,   2],
        ...     "code/n_occurrences": [4,   5,   3,   2],
        ... })
        >>> stage_cfg = DictConfig({"additional_codes": ["^BIRTH$", "^VISIT.*$"]})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (3, 3)
        ┌────────────┬──────────────────┬───────────┐
        │ subject_id ┆ code             ┆ modifier1 │
        │ ---        ┆ ---              ┆ ---       │
        │ i64        ┆ str              ┆ i64       │
        ╞════════════╪══════════════════╪═══════════╡
        │ 1          ┆ BIRTH            ┆ 1         │
        │ 1          ┆ VISIT//PULMONARY ┆ 1         │
        │ 2          ┆ BIRTH            ┆ 2         │
        └────────────┴──────────────────┴───────────┘

    This stage works even if the default row index column exists:
        >>> code_metadata_df = pl.DataFrame({
        ...     "code":               ["A", "A", "B", "C"],
        ...     "modifier1":          [1,   2,   1,   2],
        ...     "code/n_subjects":    [2,   1,   3,   2],
        ...     "code/n_occurrences": [4,   5,   3,   2],
        ... })
        >>> data = pl.DataFrame({
        ...     "subject_id": [1,   1,   2,   2],
        ...     "code":       ["A", "B", "A", "C"],
        ...     "modifier1":  [1,   1,   2,   2],
        ...     "_row_idx":   [1,   1,   1,   1],
        ... }).lazy()
        >>> stage_cfg = DictConfig({"min_subjects_per_code": 2, "min_occurrences_per_code": 3})
        >>> fn = filter_measurements_fntr(stage_cfg, code_metadata_df, ["modifier1"])
        >>> fn(data).collect()
        shape: (2, 4)
        ┌────────────┬──────┬───────────┬──────────┐
        │ subject_id ┆ code ┆ modifier1 ┆ _row_idx │
        │ ---        ┆ ---  ┆ ---       ┆ ---      │
        │ i64        ┆ str  ┆ i64       ┆ i64      │
        ╞════════════╪══════╪═══════════╪══════════╡
        │ 1          ┆ A    ┆ 1         ┆ 1        │
        │ 1          ┆ B    ┆ 1         ┆ 1        │
        └────────────┴──────┴───────────┴──────────┘
    """

    min_subjects_per_code = stage_cfg.get("min_subjects_per_code", None)
    min_occurrences_per_code = stage_cfg.get("min_occurrences_per_code", None)
    additional_codes = stage_cfg.get("additional_codes", None)

    filter_exprs = []
    if min_subjects_per_code is not None:
        filter_exprs.append(pl.col("code/n_subjects") >= min_subjects_per_code)
    if min_occurrences_per_code is not None:
        filter_exprs.append(pl.col("code/n_occurrences") >= min_occurrences_per_code)

    if additional_codes is not None:
        additional_codes_expr = pl.col("code").str.contains("|".join(additional_codes))
        if filter_exprs:
            filter_expr = pl.any_horizontal([pl.all_horizontal(filter_exprs), additional_codes_expr])
        else:
            filter_expr = additional_codes_expr

    else:
        if not filter_exprs:
            return lambda df: df
        filter_expr = pl.all_horizontal(filter_exprs)

    join_cols = ["code"]
    if code_modifiers:
        join_cols.extend(code_modifiers)

    allowed_code_metadata = (code_metadata.filter(filter_expr).select(join_cols)).lazy()

    def filter_measurements_fn(df: pl.LazyFrame) -> pl.LazyFrame:
        f"""Filters subject events to only encompass those with a set of permissible codes.

        In particular, this function filters the DataFrame to only include (code, modifier) pairs that have
        at least {min_subjects_per_code} subjects and {min_occurrences_per_code} occurrences.
        """

        idx_col = "_row_idx"
        df_columns = set(df.collect_schema().names())
        while idx_col in df_columns:
            idx_col = f"_{idx_col}"

        return (
            df.with_row_index(idx_col)
            .join(allowed_code_metadata, on=join_cols, how="inner")
            .sort(idx_col)
            .drop(idx_col)
        )

    return filter_measurements_fn

main(cfg)

TODO.

Source code in meds_torch/utils/custom_filter_measurements.py 
204
205
206
207
208
209
@hydra.main(
    version_base=None, config_path=str(PREPROCESS_CONFIG_YAML.parent), config_name=PREPROCESS_CONFIG_YAML.stem
)
def main(cfg: DictConfig):
    """TODO."""
    map_over(cfg, compute_fn=filter_measurements_fntr)