"""Count Encoder"""
from __future__ import division
import numpy as np
import pandas as pd
import category_encoders.utils as util
from copy import copy
from sklearn.base import BaseEstimator, TransformerMixin
__author__ = 'joshua t. dunn'
# COUNT_ENCODER BRANCH
[docs]class CountEncoder(BaseEstimator, TransformerMixin):
def __init__(self, verbose=0, cols=None, drop_invariant=False,
return_df=True, handle_unknown='value',
handle_missing='value',
min_group_size=None, combine_min_nan_groups=None,
min_group_name=None, normalize=False):
"""Count encoding for categorical features.
For a given categorical feature, replace the names of the groups
with the group counts.
Parameters
----------
verbose: int
integer indicating verbosity of output. 0 for none.
cols: list
a list of columns to encode, if None, all string and categorical columns
will be encoded.
drop_invariant: bool
boolean for whether or not to drop columns with 0 variance.
return_df: bool
boolean for whether to return a pandas DataFrame from transform
(otherwise it will be a numpy array).
handle_missing: str
how to handle missing values at fit time. Options are 'error', 'return_nan',
and 'value'. Default 'value', which treat NaNs as a countable category at
fit time.
handle_unknown: str, int or dict of {column : option, ...}.
how to handle unknown labels at transform time. Options are 'error'
'return_nan', 'value' and int. Defaults to None which uses NaN behaviour
specified at fit time. Passing an int will fill with this int value.
normalize: bool or dict of {column : bool, ...}.
whether to normalize the counts to the range (0, 1). See Pandas `value_counts`
for more details.
min_group_size: int, float or dict of {column : option, ...}.
the minimal count threshold of a group needed to ensure it is not
combined into a "leftovers" group. Default value is 0.01.
If float in the range (0, 1),
`min_group_size` is calculated as int(X.shape[0] * min_group_size).
Note: This value may change type based on the `normalize` variable. If True
this will become a float. If False, it will be an int.
min_group_name: None, str or dict of {column : option, ...}.
Set the name of the combined minimum groups when the defaults become
too long. Default None. In this case the category names will be joined
alphabetically with a `_` delimiter.
Note: The default name can be long and may keep changing, for example,
in cross-validation.
combine_min_nan_groups: bool or dict of {column : bool, ...}.
whether to combine the leftovers group with NaN group. Default True. Can
also be forced to combine with 'force' meaning small groups are effectively
counted as NaNs. Force can only used when 'handle_missing' is 'value' or 'error'.
Note: Will not force if it creates an binary or invariant column.
Example
-------
>>> import pandas as pd
>>> from sklearn.datasets import load_boston
>>> from category_encoders import CountEncoder
>>> bunch = load_boston()
>>> y = bunch.target
>>> X = pd.DataFrame(bunch.data, columns=bunch.feature_names)
>>> enc = CountEncoder(cols=['CHAS', 'RAD']).fit(X, y)
>>> numeric_dataset = enc.transform(X)
>>> print(numeric_dataset.info())
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 506 entries, 0 to 505
Data columns (total 13 columns):
CRIM 506 non-null float64
ZN 506 non-null float64
INDUS 506 non-null float64
CHAS 506 non-null int64
NOX 506 non-null float64
RM 506 non-null float64
AGE 506 non-null float64
DIS 506 non-null float64
RAD 506 non-null int64
TAX 506 non-null float64
PTRATIO 506 non-null float64
B 506 non-null float64
LSTAT 506 non-null float64
dtypes: float64(11), int64(2)
memory usage: 51.5 KB
None
References
----------
"""
self.return_df = return_df
self.drop_invariant = drop_invariant
self.drop_cols = []
self.verbose = verbose
self.cols = cols
self._dim = None
self.mapping = None
self.handle_unknown = handle_unknown
self.handle_missing = handle_missing
self.normalize = normalize
self.min_group_size = min_group_size
self.min_group_name = min_group_name
self.combine_min_nan_groups = combine_min_nan_groups
self.feature_names = None
self._check_set_create_attrs()
self._min_group_categories = {}
self._normalize = {}
self._min_group_name = {}
self._combine_min_nan_groups = {}
self._min_group_size = {}
self._handle_unknown = {}
self._handle_missing = {}
[docs] def fit(self, X, y=None, **kwargs):
"""Fit encoder according to X.
Parameters
----------
X : array-like, shape = [n_samples, n_features]
Training vectors, where n_samples is the number of samples
and n_features is the number of features.
y : array-like, shape = [n_samples]
Target values.
Returns
-------
self : encoder
Returns self.
"""
# first check the type
X = util.convert_input(X)
self._dim = X.shape[1]
# if columns aren't passed, just use every string column
if self.cols is None:
self.cols = util.get_obj_cols(X)
else:
self.cols = util.convert_cols_to_list(self.cols)
self._check_set_create_dict_attrs()
self._fit_count_encode(X, y)
X_temp = self.transform(X, override_return_df=True)
self.feature_names = list(X_temp.columns)
if self.drop_invariant:
self.drop_cols = []
generated_cols = util.get_generated_cols(X, X_temp, self.cols)
self.drop_cols = [
x for x in generated_cols if X_temp[x].var() <= 10e-5
]
try:
[self.feature_names.remove(x) for x in self.drop_cols]
except KeyError as e:
if self.verbose > 0:
print("Could not remove column from feature names."
"Not found in generated cols.\n{}".format(e))
return self
def _fit_count_encode(self, X_in, y):
"""Perform the count encoding."""
X = X_in.copy(deep=True)
if self.cols is None:
self.cols = X.columns.values
self.mapping = {}
for col in self.cols:
if X[col].isnull().any():
if self._handle_missing[col] == 'error':
raise ValueError(
'Missing data found in column %s at fit time.'
% (col,)
)
elif self._handle_missing[col] not in ['value', 'return_nan', 'error', None]:
raise ValueError(
'%s key in `handle_missing` should be one of: '
' `value`, `return_nan` and `error` not `%s`.'
% (col, str(self._handle_missing[col]))
)
self.mapping[col] = X[col].value_counts(
normalize=self._normalize[col],
dropna=False
)
self.mapping[col].index = self.mapping[col].index.astype(object)
if self._handle_missing[col] == 'return_nan':
self.mapping[col][np.NaN] = np.NaN
# elif self._handle_missing[col] == 'value':
#test_count.py failing self.mapping[col].loc[-2] = 0
if any([val is not None for val in self._min_group_size.values()]):
self.combine_min_categories(X)
def _transform_count_encode(self, X_in, y):
"""Perform the transform count encoding."""
X = X_in.copy(deep=True)
for col in self.cols:
X[col] = X.fillna(value=np.nan)[col]
if self._min_group_size is not None:
if col in self._min_group_categories.keys():
X[col] = (
X[col].map(self._min_group_categories[col])
.fillna(X[col])
)
X[col] = X[col].astype(object).map(self.mapping[col])
if isinstance(self._handle_unknown[col], (int, np.integer)):
X[col] = X[col].fillna(self._handle_unknown[col])
elif (self._handle_unknown[col] == 'value'
and X[col].isna().any()
and self._handle_missing[col] != 'return_nan'
):
X[col].replace(np.nan, 0, inplace=True)
elif (
self._handle_unknown[col] == 'error'
and X[col].isnull().any()
):
raise ValueError(
'Missing data found in column %s at transform time.'
% (col,)
)
return X, self.mapping
[docs] def combine_min_categories(self, X):
"""Combine small categories into a single category."""
for col, mapper in self.mapping.items():
if self._normalize[col] and isinstance(self._min_group_size[col], int):
self._min_group_size[col] = self._min_group_size[col] / X.shape[0]
elif not self._normalize and isinstance(self._min_group_size[col], float):
self._min_group_size[col] = self._min_group_size[col] * X.shape[0]
if self._combine_min_nan_groups[col] is True:
min_groups_idx = mapper < self._min_group_size[col]
elif self._combine_min_nan_groups[col] == 'force':
min_groups_idx = (
(mapper < self._min_group_size[col])
| (mapper.index.isnull())
)
else:
min_groups_idx = (
(mapper < self._min_group_size[col])
& (~mapper.index.isnull())
)
min_groups_sum = mapper.loc[min_groups_idx].sum()
if (
min_groups_sum > 0
and min_groups_idx.sum() > 1
and not min_groups_idx.loc[~min_groups_idx.index.isnull()].all()
):
if isinstance(self._min_group_name[col], str):
min_group_mapper_name = self._min_group_name[col]
else:
min_group_mapper_name = '_'.join([
str(idx)
for idx
in mapper.loc[min_groups_idx].index.astype(str).sort_values()
])
self._min_group_categories[col] = {
cat: min_group_mapper_name
for cat
in mapper.loc[min_groups_idx].index.tolist()
}
if not min_groups_idx.all():
mapper = mapper.loc[~min_groups_idx]
mapper[min_group_mapper_name] = min_groups_sum
self.mapping[col] = mapper
def _check_set_create_attrs(self):
"""Check attributes setting that don't play nicely `self.cols`."""
if not (
(self.combine_min_nan_groups in ['force', True, False, None])
or isinstance(self.combine_min_nan_groups, dict)
):
raise ValueError(
"'combine_min_nan_groups' should be one of: "
"['force', True, False, None] or type dict."
)
if (
self.handle_missing == 'return_nan'
and self.combine_min_nan_groups == 'force'
):
raise ValueError(
"Cannot have `handle_missing` == 'return_nan' and "
"'combine_min_nan_groups' == 'force' for all columns."
)
if (
self.combine_min_nan_groups is not None
and self.min_group_size is None
):
pass
# raise ValueError(
# "`combine_min_nan_groups` only works when `min_group_size` "
# "is set for all columns."
# )
if (
self.min_group_name is not None
and self.min_group_size is None
):
raise ValueError(
"`min_group_name` only works when `min_group_size` is set "
"for all columns."
)
if self.combine_min_nan_groups is None:
self.combine_min_nan_groups = True
def _check_set_create_dict_attrs(self):
"""Check attributes that can be dicts and format for all `self.cols`."""
dict_attrs = {
'normalize': False,
'min_group_name': None,
'combine_min_nan_groups': True,
'min_group_size': None,
'handle_unknown': 'value',
'handle_missing': 'value',
}
for attr_name, attr_default in dict_attrs.items():
attr = copy(getattr(self, attr_name))
if isinstance(attr, dict):
for col in self.cols:
if col not in attr:
attr[col] = attr_default
setattr(self, '_' + attr_name, attr)
else:
attr_dict = {}
for col in self.cols:
attr_dict[col] = attr
setattr(self, '_' + attr_name, attr_dict)
for col in self.cols:
if (
self._handle_missing[col] == 'return_nan'
and self._combine_min_nan_groups[col] == 'force'
):
raise ValueError(
"Cannot have `handle_missing` == 'return_nan' and "
"'combine_min_nan_groups' == 'force' for columns `%s`."
% (col,)
)
if (
self._combine_min_nan_groups[col] is not True
and self._min_group_size[col] is None
):
raise ValueError(
"`combine_min_nan_groups` only works when `min_group_size`"
"is set for column %s."
% (col,)
)
if (
self._min_group_name[col] is not None
and self._min_group_size[col] is None
):
raise ValueError(
"`min_group_name` only works when `min_group_size`"
"is set for column %s."
% (col,)
)
[docs] def get_feature_names(self):
"""
Returns the names of all transformed / added columns.
Returns
-------
feature_names: list
A list with all feature names transformed or added.
Note: potentially dropped features are not included!
"""
if not isinstance(self.feature_names, list):
raise ValueError("CountEncoder has to be fitted to return feature names.")
else:
return self.feature_names