# Copyright 2022 The SQLNet Company GmbH
#
# This file is licensed under the Elastic License 2.0 (ELv2).
# Refer to the LICENSE.txt file in the root of the repository
# for details.
#
"""
Contains custom class for handling the tables of a pipeline.
"""
from __future__ import annotations
import re
from typing import Sequence, Optional, Iterator, Union, Tuple, Callable, Any
import numpy as np
from numpy.typing import NDArray
import pandas as pd # type: ignore
from getml.utilities.formatting import _Formatter
from .table import Table
from .columns import Columns
[docs]class Tables:
"""
This container holds a pipeline's tables. These tables are build from the
columns for which importances can be calculated. The motivation behind this
container is to determine which tables are more important than others.
Tables can be accessed by name, index or with a NumPy array. The container
supports slicing and can be sorted and filtered. Further, the container
holds global methods to request tables' importances.
Note:
The container is an iterable. So, in addition to
:meth:`~getml.pipeline.Tables.filter` you can also use python list
comprehensions for filtering.
Example:
.. code-block:: python
all_my_tables = my_pipeline.tables
first_table = my_pipeline.tables[0]
all_but_last_10_tables = my_pipeline.tables[:-10]
important_tables = [table for table in my_pipeline.tables if
table.importance > 0.1]
names, importances = my_pipeline.tables.importances()
"""
def __init__(
self,
targets: Sequence[str],
columns: Columns,
data: Optional[Sequence[Table]] = None,
) -> None:
self._targets = targets
self._columns = columns
if data is not None:
self.data = data
else:
self._load_tables()
if not (targets and columns) and not data:
raise ValueError(
"Missing required arguments. Either provide `targets` & "
"`columns` or else provide `data`."
)
# ----------------------------------------------------------------
def __getitem__(
self, key: Union[str, int, slice, Union[NDArray[np.int_], NDArray[np.bool_]]]
) -> Union[Table, Tables, list[Table]]:
if not self.data:
raise AttributeError("Tables container not fully initialized.")
if isinstance(key, int):
return self.data[key]
if isinstance(key, slice):
tables_subset = self.data[key]
return self._make_tables(tables_subset)
if isinstance(key, str):
if key in self.names:
return [table for table in self.data if table.name == key][0]
raise AttributeError(f"No Table with name: {key}")
if isinstance(key, np.ndarray):
tables_subset = np.array(self.data)[key].tolist()
return self._make_tables(tables_subset)
raise TypeError(
"Columns can only be indexed by: int, slices, str or np.ndarray,"
f" not {type(key).__name__}"
)
# ----------------------------------------------------------------
def __iter__(self) -> Iterator[Table]:
yield from self.data
# ----------------------------------------------------------------
def __len__(self) -> int:
return len(self.data)
# ----------------------------------------------------------------
def __repr__(self) -> str:
return self._format()._render_string()
# ------------------------------------------------------------
def _repr_html_(self) -> str:
return self._format()._render_html()
# ----------------------------------------------------------------
def _format(self) -> _Formatter:
headers = [["name", "importance", "target", "marker"]]
rows = [
[
table.name,
table.importance,
table.target,
table.marker,
]
for table in self.data
]
return _Formatter(headers, rows)
# ----------------------------------------------------------------
def _load_tables(self) -> None:
"""
Gets tables data from columns
"""
tables = []
for table_target in self._targets:
importances: dict[str, float] = {
column.table: 0.0
for column in self._columns
if column.target == table_target
}
targets: dict[str, str] = {}
markers: dict[str, str] = {}
for column in self._columns:
if column.target == table_target:
importances[column.table] += column.importance
targets[column.table] = column.target
markers[column.table] = column.marker
tables_zip = zip(
importances.keys(),
importances.values(),
targets.values(),
markers.values(),
)
for name, importance, target, marker in tables_zip:
tables.append(
Table(
name=name, importance=importance, target=target, marker=marker
)
)
self.data = tables
# ----------------------------------------------------------------
def _make_tables(self, data: Sequence[Table]) -> Tables:
"""
A factory to construct a :class:`getml.pipeline.Tables` container
from a list of :class:`getml.pipeline.Table`s.
"""
return Tables(self._targets, self._columns, data=data)
# ----------------------------------------------------------------
[docs] def filter(self, conditional: Callable[[Table], bool]) -> Tables:
"""
Filters the tables container.
Args:
conditional (callable, optional):
A callable that evaluates to a boolean for a given item.
Return:
:class:`getml.pipeline.Tables`:
A container of filtered tables.
Example:
.. code-block:: python
important_tables = my_pipeline.table.filter(
lambda table: table.importance > 0.1)
peripheral_tables = my_pipeline.tables.filter(
lambda table: table.marker == "[PERIPHERAL]")
"""
tables_filtered = [table for table in self.data if conditional(table)]
return self._make_tables(tables_filtered)
# ----------------------------------------------------------------
[docs] def importances(
self, target_num: int = 0, sort: bool = True
) -> Tuple[NDArray[np.str_], NDArray[np.float_]]:
"""
Returns the importances of tables.
Table importances are calculated by summing up the importances of the
columns belonging to the tables. Each column is assigned an importance
value that measures its contribution to the predictive performance. For
each target, the importances add up to 1.
Args:
target_num (int):
Indicates for which target you want to view the
importances. (Pipelines can have more than one target.)
sort (bool):
Whether you want the results to be sorted.
Return:
(:class:`numpy.ndarray`, :class:`numpy.ndarray`):
- The first array contains the names of the tables.
- The second array contains their importances. By definition,
all importances add up to 1.
"""
target_name = self._targets[target_num]
names = np.empty(0, dtype=str)
importances = np.empty(0, dtype=float)
for table in self.data:
if table.target == target_name:
names = np.append(names, table.name)
importances = np.append(importances, table.importance)
if not sort:
return names, importances
indices = np.argsort(importances)[::-1]
return (names[indices], importances[indices])
# ----------------------------------------------------------------
@property
def names(self) -> list[str]:
"""
Holds the names of a :class:`~getml.Pipeline`\'s tables.
Returns:
:class:`list` containing the names.
Note:
The order corresponds to the current sorting of the container.
"""
return [table.name for table in self.data]
# ----------------------------------------------------------------
[docs] def sort(
self,
by: Optional[str] = None,
key: Optional[Callable[[Table], Any]] = None,
descending: Optional[bool] = None,
) -> Tables:
"""
Sorts the Tables container. If no arguments are provided the
container is sorted by target and name.
Args:
by (str, optional):
The name of field to sort by. Possible fields:
- name(s)
- importances(s)
key (callable, optional):
A callable that evaluates to a sort key for a given item.
descending (bool, optional):
Whether to sort in descending order.
Return:
:class:`getml.pipeline.Tables`:
A container of sorted tables.
Example:
.. code-block:: python
by_importance = my_pipeline.tables.sort(key=lambda table: table.importance)
"""
reverse = False if descending is None else descending
if (by is not None) and (key is not None):
raise ValueError("Only one of `by` and `key` can be provided.")
if key is not None:
tables_sorted = sorted(self.data, key=key, reverse=reverse)
return self._make_tables(tables_sorted)
if by is None:
tables_sorted = sorted(
self.data, key=lambda table: table.name, reverse=reverse
)
tables_sorted.sort(key=lambda table: table.target)
return self._make_tables(tables_sorted)
if re.match(pattern="names?$", string=by):
tables_sorted = sorted(
self.data, key=lambda table: table.name, reverse=reverse
)
return self._make_tables(tables_sorted)
if re.match(pattern="importances?$", string=by):
reverse = True if descending is None else descending
tables_sorted = sorted(
self.data, key=lambda table: table.importance, reverse=reverse
)
return self._make_tables(tables_sorted)
raise ValueError(f"Cannot sort by: {by}.")
# ----------------------------------------------------------------
@property
def targets(self) -> list[str]:
"""
Holds the targets of a :class:`~getml.Pipeline`\'s tables.
Returns:
:class:`list` containing the names.
Note:
The order corresponds to the current sorting of the container.
"""
return [table.target for table in self.data]
# ----------------------------------------------------------------
[docs] def to_pandas(self) -> pd.DataFrame:
"""
Returns all information related to the tables in a pandas DataFrame.
"""
data_frame = pd.DataFrame()
for i, table in enumerate(self.data):
data_frame.loc[i, "name"] = table.name
data_frame.loc[i, "importance"] = table.importance
data_frame.loc[i, "target"] = table.target
data_frame.loc[i, "marker"] = table.marker
return data_frame