category_encoders/invalid-handle-option
| Rule | invalid-handle-option |
|---|---|
| Pack | Category Encoders |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
The encoder does not take this handle_unknown or handle_missing value, so fit raises ValueError on category_encoders 2.10 and later. Use one of the values the message names, such as "value".
encoder = ce.OneHotEncoder(cols=["city"], handle_unknown="ignore")encoder = ce.OneHotEncoder(cols=["city"], handle_unknown="value")What the rule reports
Section titled “What the rule reports”The rule reports a handle_unknown or handle_missing string that the encoder doesn't take, such
as handle_unknown="ignore", which belongs to the scikit-learn OneHotEncoder.
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
The unsafe fix writes "value", the default, for each string the class doesn't take, and on HashingEncoder the keyword is dropped. Since 2.10 the call raised at fit, and before 2.10 the string did not act as the default, so the codes can change.
| Encoders | handle_unknown |
handle_missing |
|---|---|---|
The ten supervised encoders, and OrdinalEncoder |
error, return_nan, value |
error, return_nan, value |
BackwardDifferenceEncoder, BaseNEncoder, BinaryEncoder, GrayEncoder, HelmertEncoder, PolynomialEncoder, SumEncoder |
the three, and indicator |
the three, and indicator |
OneHotEncoder, MultiHotEncoder |
the three, and indicator |
the three, indicator and ignore |
RankHotEncoder |
the three | the three, and ignore |
HashingEncoder |
not a parameter | not a parameter |
CountEncoder |
not checked | not checked |
The table was read from 2.11.2 by fitting every class with every string. Since 2.10 an unknown
string raises at fit, as in ValueError: Unexpected handle_unknown value 'ignore' for OneHotEncoder. Supported values: ['error', 'indicator', 'return_nan', 'value']. Before 2.10 the
string was taken in silence and did not act as the default. On 2.9.0, handle_unknown="ignore"
coded an unseen category as NaN on OneHotEncoder, OrdinalEncoder, BinaryEncoder,
TargetEncoder and WOEEncoder. The form is wrong on both sides of 2.10, so the rule runs on
every version.
"value" on a category_encoders OneHotEncoder codes an unseen category as zeros, which is
what "ignore" does in scikit-learn. HashingEncoder keeps no list of categories, so it has
neither keyword and raised TypeError: HashingEncoder.__init__() got an unexpected keyword argument 'handle_unknown'. A value that is not a string, such as a number or a function, is
allowed since 2.11 and is left alone. CountEncoder checks nothing and is left alone too.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
$C takes "error", "return_nan" or "value" here. $KW raises ValueError at fit since category_encoders 2.10, and before that it was taken in silence and did not act as the default. Use "value", the default.Browse the other rules of the Category Encoders pack.