Make Outlier respect dtype, remove Clipping

I added support for `Outlier` to respect dtypes - series that contain
integers won't be cast to float, but preserve the int dtype.

I ended up removing the `Clipping` Error Type, because it introduces
a dependency between error_pct and the error type that the library
currently does not support. A bigger effort would be needed to
properly support `Clipping`.

I added one example of how to use `Outlier`.

error_generation/error_type/__init__.py

@@ -1,7 +1,6 @@
 from ._base import ErrorType
 from .add_delta import AddDelta
 from .butterfinger import Butterfinger
-from .clipping import Clipping
 from .extraneous import Extraneous
 from .mislabel import Mislabel
 from .missing import MissingValue

error_generation/error_type/outlier.py

@@ -3,7 +3,7 @@
 from typing import TYPE_CHECKING
 
 import numpy as np
-from pandas.api.types import is_numeric_dtype
+from pandas.api.types import is_integer_dtype, is_numeric_dtype
 
 from error_generation.error_type import ErrorType
 from error_generation.utils import get_column
@@ -54,6 +54,10 @@ def _apply(self: Outlier, table: pd.DataFrame, error_mask: pd.DataFrame, column:
         perturbation_upper = self.config.outlier_coefficient * (upper_boundary - mean_value)
         perturbation_lower = self.config.outlier_coefficient * (mean_value - lower_boundary)
 
+        if is_integer_dtype(series):  # round float to int when series is int
+            perturbation_upper = np.ceil(perturbation_upper)
+            perturbation_lower = np.floor(perturbation_lower)
+
         # Get masks for the different outlier types depending on the mean
         mask_lower = (series < mean_value) & series_mask
         mask_upper = (series > mean_value) & series_mask
@@ -73,6 +77,10 @@ def _apply(self: Outlier, table: pd.DataFrame, error_mask: pd.DataFrame, column:
 
         # Apply Gaussian noise to simulate the increase in measurement error of the outliers
         noise_std = self.config.outlier_noise_coeff * iqr
-        series.loc[series_mask] += rng.normal(loc=0, scale=noise_std, size=series_mask.sum())
+
+        if is_integer_dtype(series):  # round float to int when series is int
+            series.loc[series_mask] += np.rint(rng.normal(loc=0, scale=noise_std, size=series_mask.sum()))
+        else:
+            series.loc[series_mask] += rng.normal(loc=0, scale=noise_std, size=series_mask.sum())
 
         return series

error_generation/utils/utils.py

@@ -55,8 +55,6 @@ class ErrorTypeConfig:
         replace_what: String that the Replace Error Type replaces with replace_with.
         replace_with: String that the Replace Error Type uses to replace replace_what with. Defaults to "".
         add_delta_value: Value that is added to the value by the AddDelta Error Type.
-        clip_lower_quantile: Lower quantile of the clipping range for Clipping Error Type. Default is None, which means no lower clipping.
-        clip_upper_quantile: Upper quantile of the clipping range for Clipping Error Type. Default is None, which means no upper clipping.
         outlier_coefficient: Coefficient that determines the magnitude of the outliers for the Outlier Error Type.
         outlier_noise_coeff: Coefficient that influences the standard deviation of the noise added to the outliers for the Outlier Error Type.
     """
@@ -87,9 +85,6 @@ class ErrorTypeConfig:
 
     add_delta_value: Any | None = None
 
-    clip_lower_quantile: float | None = None
-    clip_upper_quantile: float | None = None
-
     outlier_coefficient: float = 1.0
     outlier_noise_coeff: float = 0.1