Linear Model

Inspector functionality specific to linear regression, logistic regression, and their variants

Linear Regression

import sklearn.datasets
from model_inspector import get_inspector
from sklearn.linear_model import Ridge
X, y = sklearn.datasets.load_diabetes(return_X_y=True, as_frame=True)

inspector = get_inspector(Ridge().fit(X, y), X, y)

source

plot_coefs_vs_hparam (regression model)

 plot_coefs_vs_hparam (regression model) (hparam:str,
                       vals:Sequence[float])

Plot coefficient values against a hyperparameter

Parameters:

  • hparam: Name of hyperparameter; must be an attribute of self.model
  • vals: Values of that hyperparameter to use
ax = inspector.plot_coefs_vs_hparam("alpha", np.logspace(-2, 0.5, 10))

source

plot_waterfall (regression model)

 plot_waterfall (regression model)
                 (item:Union[pandas.core.series.Series,<built-
                 infunctionarray>], bar_num_formatter:str='.1f',
                 tick_num_formatter:str='.2f', sorted_value=True,
                 threshold=0.01, blue_color='#377eb8',
                 green_color='#4daf4a', red_color='#ff7f00', Title='',
                 x_lab='', y_lab='', formatting='{:,.1f}',
                 other_label='other', net_label='net', rotation_value=30)

Make a waterfall chart showing how each feature contributes to the prediction for the input item.

Parameters:

  • item: Input item, with the same shape and value meanings as a single row from self.X
  • bar_num_formatter: Bar label format specifier
  • tick_num_formatter: Tick label format specifier

Additional keyword arguments will be passed to waterfall_chart.plot

Remaining parameters are passed to waterfall_chart.plot.

ax = inspector.plot_waterfall(X.iloc[0])

source

show_model (regression model)

 show_model (regression model) (intercept_formatter:str='.2f',
             coef_formatter:str='.2f')

Show model equation

Parameters:

  • intercept_formatter: Intercept format specifier
  • coef_formatter: Intercept format specifier
inspector.show_model()
target = 152.13 + 29.47 * age - 83.15 * sex + 306.35 * bmi + 201.63 * bp + 5.91 * s1 - 29.52 * s2 - 152.04 * s3 + 117.31 * s4 + 262.94 * s5 + 111.88 * s6

Binary Logistic Regression

from sklearn.linear_model import LogisticRegression
X, y = sklearn.datasets.load_breast_cancer(return_X_y=True, as_frame=True)
X = X.iloc[:, :10]
inspector = get_inspector(LogisticRegression(max_iter=1_000).fit(X, y), X, y)

source

plot_coefs_vs_hparam (binary model)

 plot_coefs_vs_hparam (binary model) (hparam:str, vals:Sequence[float])

Plot coefficient values against a hyperparameter

Parameters:

  • hparam: Name of hyperparameter; must be an attribute of self.model
  • vals: Values of that hyperparameter to use
ax = inspector.plot_coefs_vs_hparam("C", np.logspace(-2, 0.5, 10))

source

show_model (binary model)

 show_model (binary model) (intercept_formatter:str='.2f',
             coef_formatter:str='.2f')

Show model equation

Parameters:

  • intercept_formatter: Intercept format specifier
  • coef_formatter: Intercept format specifier
inspector.show_model()
log-odds(target) = 21.27 + 2.69 * mean radius - 0.23 * mean texture - 0.61 * mean perimeter + 0.00 * mean area - 0.48 * mean smoothness - 0.79 * mean compactness - 1.42 * mean concavity - 0.76 * mean concave points - 0.69 * mean symmetry - 0.12 * mean fractal dimension

source

plot_waterfall (binary model)

 plot_waterfall (binary model)
                 (item:Union[pandas.core.series.Series,<built-
                 infunctionarray>], bar_num_formatter:str='.1f',
                 tick_num_formatter:str='.2f', sorted_value=True,
                 threshold=0.01, blue_color='#377eb8',
                 green_color='#4daf4a', red_color='#ff7f00', Title='',
                 x_lab='', y_lab='', formatting='{:,.1f}',
                 other_label='other', net_label='net', rotation_value=30)

Make a waterfall chart showing how each feature contributes to the prediction for the input item for a binary classification model.

Parameters:

  • item: Input item, with the same shape and value meanings as a single row from self.X
  • bar_num_formatter: Bar label format specifier
  • tick_num_formatter: Tick label format specifier
  • waterfall_kwargs: kwargs to pass to waterfall_chart.plot

Remaining parameters are passed to waterfall_chart.plot.

ax = inspector.plot_waterfall(X.iloc[0])

Multiclass Logistic Regression

from sklearn.linear_model import LogisticRegression
X, y = sklearn.datasets.load_iris(return_X_y=True, as_frame=True)
inspector = get_inspector(LogisticRegression(max_iter=10_000).fit(X, y), X, y)

source

plot_coefs_vs_hparam (multiclass model)

 plot_coefs_vs_hparam (multiclass model) (hparam:str,
                       vals:Sequence[float])

Plot coefficient values against a hyperparameter

Parameters:

  • hparam: Name of hyperparameter; must be an attribute of self.model
  • vals: Values of that hyperparameter to use

Returns NumPy array of Axes objects.

axes = inspector.plot_coefs_vs_hparam("C", np.logspace(-2, 0.5, 10))

source

show_model (multiclass model)

 show_model (multiclass model) (intercept_formatter:str='.2f',
             coef_formatter:str='.2f')

Show model equation

Parameters:

  • intercept_formatter: Intercept format specifier
  • coef_formatter: Intercept format specifier
inspector.show_model()

log-odds(target = 0) = 9.85 - 0.42 * sepal length (cm) + 0.97 * sepal width (cm) - 2.52 * petal length (cm) - 1.08 * petal width (cm)

log-odds(target = 1) = 2.24 + 0.53 * sepal length (cm) - 0.32 * sepal width (cm) - 0.21 * petal length (cm) - 0.94 * petal width (cm)

log-odds(target = 2) = -12.09 - 0.11 * sepal length (cm) - 0.65 * sepal width (cm) + 2.72 * petal length (cm) + 2.02 * petal width (cm)