Regression¶

Available Models¶

logsuite provides 6 regression models for crossplot analysis:

Model	Equation	Use Case
`LinearRegression`	y = ax + b	General linear trends
`PolynomialRegression`	y = a₀ + a₁x + … + aₙxⁿ	Non-linear trends
`ExponentialRegression`	y = a·eᵇˣ	Permeability vs porosity
`LogarithmicRegression`	y = a·ln(x) + b	Diminishing returns
`PowerRegression`	y = a·xᵇ	Power law relationships
`PolynomialExponentialRegression`	y = exp(a₀ + a₁x + … + aₙxⁿ)	Complex perm-poro

Basic Usage¶

from logsuite import ExponentialRegression
import numpy as np

x = well.PHIE.values
y = well.PERM.values

# Fit regression
reg = ExponentialRegression()
reg.fit(x, y)

# Get equation and R²
print(reg.equation)
print(f"R² = {reg.r_squared:.4f}")

# Predict
y_pred = reg.predict(np.linspace(0.05, 0.35, 100))

With Crossplots¶

from logsuite import Crossplot

xplot = Crossplot(manager, x="PHIE", y="PERM")
xplot.add_regression("polynomial_2")
xplot.show()

Regression on a subset¶

add_regression accepts a where= argument — a dict of public column names to allowed values, or a callable returning a boolean mask. Subsets smaller than min_samples (default 5) are skipped with a warning:

xplot.add_regression("exponential", name="Sand 2",
                     where={"Facies": ["Reservoir"]})

For “one regression per category” use the convenience method:

xplot.add_regression_per("Facies", "exponential")

See the per-group regression how-to.

Polynomial Degree¶

For PolynomialRegression and PolynomialExponentialRegression:

reg = PolynomialRegression(degree=3)
reg.fit(x, y)

Petrel calculator syntax¶

Each regression artifact exposes its equation in three forms — natural, log10, and Petrel:

fit = manager.properties(["PHIE", "PERM"]).fit(
    ExponentialRegression(), name="all wells", equation_format="petrel"
)
fit.equation(format="petrel")    # 'pow(10, 19.628*x - 3.676)' — paste into Petrel

See the Petrel-export how-to.