Implement utilities for SHAP values in regression

This commit is contained in:
RunasSudo 2023-02-07 18:50:07 +11:00
parent dbebc3b8e9
commit 967b853b02
Signed by: RunasSudo
GPG Key ID: 7234E476BF21C61A
2 changed files with 81 additions and 0 deletions

View File

@ -32,6 +32,7 @@ import weakref
from .bayes_factors import BayesFactor, bayesfactor_afbf
from .config import config
from .shap import ShapResult
from .sig_tests import ChiSquaredResult, FTestResult
from .utils import Estimate, PValueStyle, as_numeric, check_nan, cols_for_formula, convert_pandas_nullable, fmt_p, formula_factor_ref_category, parse_patsy_term
@ -460,6 +461,26 @@ class RegressionResult:
self.exp
)
def shap(self, **kwargs):
# TODO: Documentation
import shap
xdata = ShapResult._get_xdata(self)
# Combine terms into single list
params = []
for term in self.terms.values():
if isinstance(term, SingleTerm):
params.append(term.beta.point)
else:
params.extend(s.beta.point for s in term.categories.values())
explainer = shap.LinearExplainer((np.array(params[1:]), params[0]), xdata, **kwargs) # FIXME: Assumes zeroth term is intercept
shap_values = explainer.shap_values(xdata).astype('float')
return ShapResult(weakref.ref(self), shap_values, list(xdata.columns))
def _header_table(self, html):
"""Return the entries for the header table"""

60
yli/shap.py Normal file
View File

@ -0,0 +1,60 @@
import pandas as pd
import patsy
from .utils import as_numeric, check_nan, cols_for_formula, convert_pandas_nullable
class ShapResult:
# TODO: Documentation
def __init__(self, model, shap_values, features):
self.model = model
self.shap_values = shap_values
self.features = features
@staticmethod
def _get_xdata(model):
df = model.df()
if df is None:
raise Exception('Referenced DataFrame has been dropped')
dep = model.dep
# Check for/clean NaNs
# NaN warning/error will already have been handled in regress, so here we pass nan_policy='omit'
# Following this, we pass nan_policy='raise' to assert no NaNs remaining
df = df[[dep] + cols_for_formula(model.formula, df)]
df = check_nan(df, 'omit')
# Ensure numeric type for dependent variable
df[dep], dep_categories = as_numeric(df[dep])
# Convert pandas nullable types for independent variables as this breaks statsmodels
df = convert_pandas_nullable(df)
# Get xdata for SHAP
dmatrix = patsy.dmatrix(model.formula, df, return_type='dataframe')
xdata = dmatrix.iloc[:, 1:] # FIXME: Assumes zeroth term is intercept
return xdata
def mean(self):
return pd.Series(abs(self.shap_values).mean(axis=0), index=self.features)
def plot(self, **kwargs):
import matplotlib.pyplot as plt
import shap
model = self.model()
if model is None:
raise Exception('Referenced RegressionResult has been dropped')
xdata = self._get_xdata(model)
shap.summary_plot(self.shap_values, xdata, show=False, axis_color='black', **kwargs) # pass show=False to get gcf/gca
# Fix colour bar
# https://stackoverflow.com/questions/70461753/shap-the-color-bar-is-not-displayed-in-the-summary-plot
ax_colorbar = plt.gcf().axes[-1]
ax_colorbar.set_aspect('auto')
ax_colorbar.set_box_aspect(50)
return plt.gcf(), plt.gca()