[ENH] Bayesian Linear Regression using Normal Conjugate Prior

docs/source/api_reference/regression.rst

@@ -195,3 +195,15 @@ Base classes
     :template: class.rst
 
     BaseProbaRegressor
+
+Bayesian
+--------
+
+.. currentmodule:: skpro.regression.bayesian
+
+.. autosummary::
+    :toctree: auto_generated/
+    :template: class.rst
+
+    BayesianConjugateLinearRegressor
+    BayesianLinearRegressor

examples/bayesian/test_data.csv

@@ -0,0 +1,6 @@
+,x1,x2,y_true,y_train
+11,0.9699098521619943,0.3046137691733707,3.8536610118441006,3.6611198716359423
+16,0.3042422429595377,0.4951769101112702,3.094015216252886,2.674406454641567
+1,0.9507143064099162,0.19967378215835974,3.5004499592949117,3.5861340998898967
+5,0.15599452033620265,0.6075448519014384,3.1346235963767204,2.774701492179366
+3,0.5986584841970366,0.5924145688620425,3.9745606749802005,3.824008827185556

examples/bayesian/train_data.csv

@@ -0,0 +1,21 @@
+,x1,x2,y_true,y_train
+17,0.5247564316322378,0.034388521115218396,2.152678426610131,1.9980722386845238
+7,0.8661761457749352,0.06505159298527952,2.927507070505709,3.4560681836151668
+12,0.8324426408004217,0.09767211400638387,2.9579016236199953,2.619440623467016
+18,0.43194501864211576,0.9093204020787821,4.591851243520578,4.75748295922236
+15,0.18340450985343382,0.12203823484477883,1.732923724241204,2.1985637837993033
+8,0.6011150117432088,0.9488855372533332,5.048886635246418,5.220695780030648
+10,0.020584494295802447,0.8083973481164611,3.466361032940988,3.6284030176383855
+20,0.6118528947223795,0.662522284353982,4.211272642506705,3.97168552358406
+19,0.2912291401980419,0.2587799816000169,2.358798225196135,2.8465707887573144
+6,0.05808361216819946,0.17052412368729153,1.6277395953982734,1.3974202099183797
+9,0.7080725777960455,0.9656320330745594,5.313041254815769,4.431521177134401
+21,0.13949386065204183,0.31171107608941095,2.2141209495723166,2.121291461240408
+23,0.3663618432936917,0.5467102793432796,3.3728545246172223,2.774751212576887
+22,0.29214464853521815,0.5200680211778108,3.1444933606038687,2.5913258736008546
+13,0.21233911067827616,0.6842330265121569,3.477377300893023,3.783215445313457
+14,0.18182496720710062,0.4401524937396013,2.6841074156330054,3.199607176880981
+4,0.15601864044243652,0.046450412719997725,1.4513885190448663,0.7121275238611526
+24,0.45606998421703593,0.18485445552552704,2.466703335010653,2.872966246207752
+0,0.3745401188473625,0.7851759613930136,4.104608121873766,4.473841411871471
+2,0.7319939418114051,0.5142344384136116,4.0066911988636456,3.9488670576695255

examples/bayesian/utils.py

@@ -0,0 +1,76 @@
+"""Utility function for Bayesian example notebooks."""
+
+import pandas as pd
+
+
+def style_data(data, vmax=None, subset=None, cmap="coolwarm", hide_index=False):
+    """
+    Apply styling to a Series or DataFrame.
+
+    Parameters
+    ----------
+    data : pd.DataFrame or pd.Series
+        The data to style.
+    vmax : float, optional
+        The maximum numeric value for the color spectrum.
+        Defaults to the max value in the data.
+    subset : list, optional
+        List of columns to which the formatting is to be applied (for DataFrames).
+    cmap : str, optional
+        The color map to apply for the gradient. Defaults to 'coolwarm'.
+    hide_index : bool, optional
+        If True, hide the index in the output. Defaults to False.
+
+    Returns
+    -------
+    pd.io.formats.style.Styler
+        The styled Series or DataFrame.
+    """
+    # Check if input is a Series or DataFrame
+    if isinstance(data, pd.Series):
+        # For Series, directly apply background gradient and format
+        if vmax is None:
+            vmax = data.max()
+        styled_data = (
+            data.to_frame()
+            .style.background_gradient(cmap=cmap, axis=0, vmin=-vmax, vmax=vmax)
+            .format("{:.3f}")
+        )
+
+        # Hide the index if requested
+        if hide_index:
+            styled_data = styled_data.hide(axis="index")
+
+    elif isinstance(data, pd.DataFrame):
+        # Determine the max value for the gradient if not provided
+        if vmax is None:
+            vmax = data.select_dtypes(include=["number"]).max().max()
+
+        # If no subset provided, apply to all float columns by default
+        if subset is None:
+            subset = pd.IndexSlice[:, data.select_dtypes(include=["float64"]).columns]
+
+        # Apply background gradient to numeric columns and format to 3 decimal points
+        styled_data = data.style.background_gradient(
+            cmap=cmap, axis=None, vmin=-vmax, vmax=vmax, subset=subset
+        ).format("{:.3f}", subset=subset)
+
+        # Color boolean columns (pink for False, lightblue for True)
+        bool_columns = data.select_dtypes(include=["bool"]).columns
+
+        def color_boolean(val):
+            color = "lightblue" if val else "pink"
+            return f"background-color: {color}"
+
+        # Apply the boolean-specific styling if any boolean columns exist
+        if not bool_columns.empty:
+            styled_data = styled_data.applymap(color_boolean, subset=bool_columns)
+
+        # Hide the index if requested
+        if hide_index:
+            styled_data = styled_data.hide(axis="index")
+
+    else:
+        raise TypeError("Input must be a pandas DataFrame or Series.")
+
+    return styled_data

skpro/regression/bayesian/__init__.py

@@ -0,0 +1,8 @@
+"""Base classes for Bayesian probabilistic regression."""
+# copyright: skpro developers, BSD-3-Clause License (see LICENSE file)
+
+__all__ = ["BayesianConjugateLinearRegressor"]
+
+from skpro.regression.bayesian.bayesian_conjugate import (
+    BayesianConjugateLinearRegressor,
+)