MultilabelBinaryMetric(
source_col,
target_cols_prefix,
positive_indexes,
lower_thres=0.5,
function="accuracy",
name_suffix="",
)
Bases: DfMetric
Converts a multilabel classification problem to a binary.
Uses the positive_indexes class as positive class.
If any of these class indexes is higher than threshold
it is counted as positive class and negative otherwise.
Parameters
source_col: str
name of the column with the class index or indexes (multitarget)
target_cols_prefix: str
starting name of the columns containing the output of the model
positive_indexes: List[int]
indexes of the positive classes.
Has to be match with the index in classes in the
datasets.yml -> data_description
lower_thres: float, default 0.5
lower threshold for the positive classes to be counted as true
function: str or dict, default "accuracy"
function to calculate the metric.
If str is given it must be a key of the cl_metric_dict.
Otherwise the given dict is initialized with init_object
and additional the key name
must be available to define the output
name in the result dict.
name_suffix: str, default ""
can be used to add a str to the dict output name
(e.g. if the same metric should be applied
to multiple columns)
Source code in niceml/mlcomponents/resultanalyzers/dataframes/multilabeltobinary.py
| def __init__(
self,
source_col: str,
target_cols_prefix: str,
positive_indexes: List[int],
lower_thres: float = 0.5,
function: Union[str, dict] = "accuracy",
name_suffix: str = "",
):
"""
Converts a multilabel classification problem to a binary.
Uses the `positive_indexes` class as positive class.
If any of these class indexes is higher than `threshold`
it is counted as positive class and negative otherwise.
Parameters
----------
source_col: str
name of the column with the class index or indexes (multitarget)
target_cols_prefix: str
starting name of the columns containing the output of the model
positive_indexes: List[int]
indexes of the positive classes.
Has to be match with the index in `classes` in the
datasets.yml -> data_description
lower_thres: float, default 0.5
lower threshold for the positive classes to be counted as true
function: str or dict, default "accuracy"
function to calculate the metric.
If str is given it must be a key of the `cl_metric_dict`.
Otherwise the given dict is initialized with `init_object`
and additional the key `name`
must be available to define the output
name in the result dict.
name_suffix: str, default ""
can be used to add a str to the dict output name
(e.g. if the same metric should be applied
to multiple columns)
"""
self.source_col = source_col
self.target_cols_prefix = target_cols_prefix
self.name_suffix = name_suffix
self.positive_indexes = positive_indexes
self.lower_thres = lower_thres
if type(function) is str:
self.func_name = function
try:
self.function = cl_metric_dict[function]
except KeyError as e:
raise Exception(
f"Function with name {function} not supported!"
f" Available: {list(cl_metric_dict.keys())}"
) from e
else:
self.func_name = function["name"]
self.function = function["target"]
|
Functions