Skip to content

eds_scikit.io.hive

HiveData

HiveData(database_name: str, spark_session: Optional[SparkSession] = None, person_ids: Optional[Iterable[int]] = None, tables_to_load: Optional[Union[Dict[str, Optional[List[str]]], List[str]]] = None, columns_to_load: Optional[Union[Dict[str, Optional[List[str]]], List[str]]] = None, database_type: Optional[str] = 'OMOP', prune_omop_date_columns: bool = True, cache: bool = True)

Bases: BaseData

Spark interface for OMOP data stored in a Hive database.

This class provides a simple access to data stored in Hive. Data is returned as koalas dataframes that match the tables stored in Hive.

PARAMETER DESCRIPTION
database_name

The name of you database in Hive. Ex: "cse_82727572"

TYPE: str

spark_session

If None, a SparkSession will be retrieved or created via SparkSession.builder.enableHiveSupport().getOrCreate()

TYPE: pyspark.sql.SparkSession DEFAULT: None

person_ids

An iterable of person_id that is used to define a subset of the database.

TYPE: Optional[Iterable[int]] DEFAULT: None

tables_to_load

deprecated

TYPE: dict, default DEFAULT: None

columns_to_load

deprecated

TYPE: dict, default DEFAULT: None

database_type

Whether to use the native OMOP schema or to convert I2B2 inputs to OMOP.

TYPE: Optional[str] DEFAULT: 'OMOP'

prune_omop_date_columns

In OMOP, most date values are stored both in a <str>_date and <str>_datetime column Koalas has trouble handling the date time, so we only keep the datetime column

TYPE: bool DEFAULT: True

cache

Whether to cache each table after preprocessing or not. Will speed-up subsequent calculations, but can be long/infeasable for very large tables

TYPE: bool DEFAULT: True

ATTRIBUTE DESCRIPTION
person

Hive data for table person as a koalas dataframe. Other OMOP tables can also be accessed as attributes

TYPE: koalas dataframe

available_tables

names of OMOP tables that can be accessed as attributes with this HiveData object.

TYPE: list of str

Examples:

data = HiveData(database_name="edsomop_prod_a")
data.available_tables
# Out: ["person", "care_site", "condition_occurrence", ... ]

person = data.person
type(person)
# Out: databricks.koalas.frame.DataFrame

person["person_id"].count()
# Out: 12670874

This class can be used to create a subset of data for a given list of person_id. This is useful because the smaller dataset can then be used to prototype more rapidly.

my_person_ids = [9226726, 2092082, ...]
data = HiveData(
    spark_session=spark, database_name="edsomop_prod_a", person_ids=my_person_ids
)
data.person["person_id"].count()
# Out: 1000

tables_to_save = ["person", "visit_occurrence"]
data.persist_tables_to_folder("./cohort_sample_1000", table_names=tables_to_save)
# Out: writing /export/home/USER/cohort_sample_1000/person.parquet
# Out: writing /export/home/USER/cohort_sample_1000/visit_occurrence.parquet
# Out: ...
Source code in eds_scikit/io/hive.py
 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
def __init__(
    self,
    database_name: str,
    spark_session: Optional[SparkSession] = None,
    person_ids: Optional[Iterable[int]] = None,
    tables_to_load: Optional[
        Union[Dict[str, Optional[List[str]]], List[str]]
    ] = None,
    columns_to_load: Optional[
        Union[Dict[str, Optional[List[str]]], List[str]]
    ] = None,
    database_type: Optional[str] = "OMOP",
    prune_omop_date_columns: bool = True,
    cache: bool = True,
):
    """Spark interface for OMOP data stored in a Hive database.

    This class provides a simple access to data stored in Hive.
    Data is returned as koalas dataframes that match the tables
    stored in Hive.

    Parameters
    ----------
    database_name : str
        The name of you database in Hive. Ex: "cse_82727572"
    spark_session : pyspark.sql.SparkSession
        If None, a SparkSession will be retrieved or  created via `SparkSession.builder.enableHiveSupport().getOrCreate()`
    person_ids : Optional[Iterable[int]]
        An iterable of `person_id` that is used to define a subset of the database.
    tables_to_load : dict, default=None
        *deprecated*
    columns_to_load : dict, default=None
        *deprecated*
    database_type: Optional[str] = 'OMOP'. Must be 'OMOP' or 'I2B2'
        Whether to use the native OMOP schema or to convert I2B2 inputs to OMOP.
    prune_omop_date_columns: bool, default=True
        In OMOP, most date values are stored both in a `<str>_date` and `<str>_datetime` column
        Koalas has trouble handling the `date` time, so we only keep the `datetime` column
    cache: bool, default=True
        Whether to cache each table after preprocessing or not.
        Will speed-up subsequent calculations, but can be long/infeasable for very large tables

    Attributes
    ----------
    person : koalas dataframe
        Hive data for table `person` as a koalas dataframe.
        Other OMOP tables can also be accessed as attributes
    available_tables : list of str
        names of OMOP tables that can be accessed as attributes with this
        HiveData object.

    Examples
    --------

    ```python
    data = HiveData(database_name="edsomop_prod_a")
    data.available_tables
    # Out: ["person", "care_site", "condition_occurrence", ... ]

    person = data.person
    type(person)
    # Out: databricks.koalas.frame.DataFrame

    person["person_id"].count()
    # Out: 12670874
    ```

    This class can be used to create a subset of data for a given
    list of `person_id`. This is useful because the smaller dataset
    can then be used to prototype more rapidly.

    ```python
    my_person_ids = [9226726, 2092082, ...]
    data = HiveData(
        spark_session=spark, database_name="edsomop_prod_a", person_ids=my_person_ids
    )
    data.person["person_id"].count()
    # Out: 1000

    tables_to_save = ["person", "visit_occurrence"]
    data.persist_tables_to_folder("./cohort_sample_1000", table_names=tables_to_save)
    # Out: writing /export/home/USER/cohort_sample_1000/person.parquet
    # Out: writing /export/home/USER/cohort_sample_1000/visit_occurrence.parquet
    # Out: ...
    ```

    """
    super().__init__()

    if columns_to_load is not None:
        logger.warning("'columns_to_load' is deprecated and won't be used")

    if tables_to_load is not None:
        logger.warning("'tables_to_load' is deprecated and won't be used")

    self.spark_session = (
        spark_session or SparkSession.builder.enableHiveSupport().getOrCreate()
    )
    self.database_name = database_name
    if database_type not in ["I2B2", "OMOP"]:
        raise ValueError(
            f"`database_type` must be either 'I2B2' or 'OMOP'. Got {database_type}"
        )
    self.database_type = database_type

    if self.database_type == "I2B2":
        self.database_source = "cse" if "cse" in self.database_name else "edsprod"
        self.omop_to_i2b2 = settings.i2b2_tables[self.database_source]
        self.i2b2_to_omop = defaultdict(list)
        for omop_table, i2b2_table in self.omop_to_i2b2.items():
            self.i2b2_to_omop[i2b2_table].append(omop_table)

    self.prune_omop_date_columns = prune_omop_date_columns
    self.cache = cache
    self.user = os.environ["USER"]
    self.person_ids, self.person_ids_df = self._prepare_person_ids(person_ids)
    self.available_tables = self.list_available_tables()
    self._tables = {}

persist_tables_to_folder

persist_tables_to_folder(folder: str, person_ids: Optional[Iterable[int]] = None, tables: List[str] = None, overwrite: bool = False) -> None

Save OMOP tables as parquet files in a given folder.

PARAMETER DESCRIPTION
folder

path to folder where the tables will be written.

TYPE: str

person_ids : iterable person_ids to keep in the subcohort.

tables : list of str, default None list of table names to save. Default value is 🇵🇾data:~eds_scikit.io.settings.default_tables_to_save.

overwrite : bool, default=False whether to overwrite files if 'folder' already exists.

Source code in eds_scikit/io/hive.py
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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
def persist_tables_to_folder(
    self,
    folder: str,
    person_ids: Optional[Iterable[int]] = None,
    tables: List[str] = None,
    overwrite: bool = False,
) -> None:
    """Save OMOP tables as parquet files in a given folder.

    Parameters
    ----------
    folder : str
        path to folder where the tables will be written.

    person_ids : iterable
        person_ids to keep in the subcohort.

    tables : list of str, default None
        list of table names to save. Default value is
        :py:data:`~eds_scikit.io.settings.default_tables_to_save`.

    overwrite : bool, default=False
        whether to overwrite files if 'folder' already exists.

    """
    # Manage tables
    if tables is None:
        tables = settings.default_tables_to_save

    unknown_tables = [
        table for table in tables if table not in self.available_tables
    ]
    if unknown_tables:
        raise ValueError(
            f"The following tables are not available : {str(unknown_tables)}"
        )

    # Create folder
    folder = Path(folder).absolute()

    if folder.exists() and overwrite:
        shutil.rmtree(folder)

    folder.mkdir(parents=True, mode=0o766)

    assert os.path.exists(folder) and os.path.isdir(
        folder
    ), f"Folder {folder} not found."

    # TODO: remove everything in this folder that is a valid
    # omop table. This prevents a user from having a
    # folder containing datasets generated from different
    # patient subsets.

    # TODO: maybe check how much the user wants to persist
    # to disk. Set a limit on the number of patients in the cohort ?

    if person_ids is not None:
        person_ids = self._prepare_person_ids(person_ids, return_df=False)

    database_path = self.get_db_path()

    for idx, table in enumerate(tables):
        if self.database_type == "I2B2":
            table_path = self._hdfs_write_orc_to_parquet(
                table, person_ids, overwrite
            )
        else:
            table_path = os.path.join(database_path, table)

        df = self.get_table_from_parquet(table_path, person_ids=person_ids)

        local_file_path = os.path.join(folder, f"{table}.parquet")
        df.to_parquet(
            local_file_path,
            allow_truncated_timestamps=True,
            coerce_timestamps="ms",
        )
        logger.info(
            f"({idx+1}/{len(tables)}) Table {table} saved at "
            f"{local_file_path} (N={len(df)})."
        )

get_db_path

get_db_path()

Get the HDFS path of the database

Source code in eds_scikit/io/hive.py
374
375
376
377
378
379
380
381
def get_db_path(self):
    """Get the HDFS path of the database"""
    return (
        self.spark_session.sql(f"DESCRIBE DATABASE EXTENDED {self.database_name}")
        .filter("database_description_item=='Location'")
        .collect()[0]
        .database_description_value
    )
Back to top