category_encoders/removed-in-category-encoders
| Rule | removed-in-category-encoders |
|---|---|
| Pack | Category Encoders |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | Category Encoders 2.0 and later |
These forms were removed from category_encoders, so the call raises on 2.11.2. Use the replacement that the message names.
encoder = ce.OneHotEncoder(impute_missing=True)encoder = ce.OneHotEncoder(handle_missing="value")What the rule reports
Section titled “What the rule reports”The rule reports forms that a release of category_encoders removed and that raise on 2.11.2.
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 safe fix writes list(...) around feature_names_out_ and get_feature_names_out() before index and count, which gave the same values on 2.6.4 and on 2.6.0, where the names were a list. The unsafe fix writes the two renamed names in category_encoders.utils, because the old mixin is another class and get_categorical_cols also names category and string columns, handle_missing="value" for impute_missing=True in a call without handle_missing, which no release took beside impute_missing, and dropping randomized= from a LeaveOneOutEncoder that has sigma or says randomized=False, because on 1.3.0 sigma alone added no noise. The other forms have no fix, and remove, append, extend, insert and pop have none because on a copy of the names they would change nothing that the code reads afterwards.
| Old form | Removed in | On 2.11.2 | Write instead |
|---|---|---|---|
impute_missing= on any encoder |
2.0 | TypeError | handle_missing="value", the default |
LeaveOneOutEncoder(randomized=True, sigma=0.05) |
2.0 | TypeError | LeaveOneOutEncoder(sigma=0.05) |
utils.get_obj_cols(df) |
2.6.3 | AttributeError | utils.get_categorical_cols(df) |
utils.TransformerWithTargetMixin |
2.8 | AttributeError | utils.SupervisedTransformerMixin |
encoded.drop(columns=["intercept"]) after a contrast encoder |
2.7 | KeyError | remove the drop |
encoder.feature_names_out_.index("a") |
2.7 | AttributeError | list(encoder.feature_names_out_).index("a") |
encoder.get_feature_names_out().index("a") |
2.6.1 | AttributeError | list(encoder.get_feature_names_out()).index("a") |
Until 2.6 the four contrast encoders, BackwardDifferenceEncoder, HelmertEncoder,
PolynomialEncoder and SumEncoder, added a column named intercept that holds 1 on every
row. 2.7 stopped adding it, so a drop of the column raises KeyError: "['intercept'] not found in axis". The rule reports the drop only in a file that names one of the four classes, and it
leaves the following drops alone:
- a drop with
errors="ignore" - a drop on a frame that the file gives a column named
interceptitself, withframe["intercept"] = 1.0orframe.insert - every drop in a file that calls
assign(intercept=...)
The last two drops take out the file's own column.
feature_names_out_ was a list until 2.6 and is a NumPy array since 2.7. The result of
get_feature_names_out is a NumPy array since 2.6.1. index, remove, append, extend,
insert, count and pop raise AttributeError on an array. feature_names_in_ is still a
list.
get_obj_cols is still in the wheels of 2.6.0, 2.6.1 and 2.6.2 and went away in 2.6.3, so its
document carries since: "2.6.3".
The fixes are unsafe, because each turns code that raises into code that runs. get_obj_cols
named the columns of the object dtype, and get_categorical_cols also names the columns of the category and string dtypes. A file that
defines its own get_obj_cols is left alone.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
impute_missing was removed in category_encoders 2.0, and $C raises TypeError on it. Use handle_missing="value", the default, in place of impute_missing=True, or handle_missing="return_nan" to keep the missing value.Browse the other rules of the Category Encoders pack.