Skip to content

GLM (Generalized Linear Model)

TL;DR

Assinatura: factory.create("glm", data=df, dependent_var="outcome", independent_vars=[...]) O que faz: Regressão linear generalizada para desfechos contínuos. Quando usar: Modelar relações lineares com diferentes distribuições (Gaussian, Gamma, Tweedie).

This page provides a quickstart reference for the RAPID_GLM pipeline, covering initialisation, fitting, and results display. For full parameter documentation, assumption details, and statistical methodology, see the GLM User Guide.

Quick Reference

Method Description Example
create() Initialize GLM factory.create("glm", data=df, dependent_var="y", independent_vars=["x1", "x2"])
fit() Train model model.fit(labels={"x1": "Predictor 1"}, cross_val=True)
summary() Display results model.summary(plots=["forest_plot", "qq_plot"])

Parameters - create()

Parameter Type Default Notes Methodological Stage
data DataFrame required Dataset Preprocessing
dependent_var str None Outcome variable Modeling
independent_vars list None Predictors Modeling
formula str None Alternative: "y ~ x1 + x2" Modeling
family str "gaussian" "gaussian", "gamma", "inv_gaussian", "tweedie" Modeling
link str "identity" "identity", "log", "inverse", "sqrt" Modeling

Parameters - fit()

Parameter Type Default Description Methodological Stage
labels dict None Readable names: {"age": "Age (years)"} Evaluation
cross_val bool True Enable cross-validation Validation
n_splits int 5 Number of folds Validation

Parameters - summary()

Parameter Type Default Common Values Methodological Stage
assumptions str/list None "all", ["Durbin-Watson", "VIF"] Evaluation
performance str/list None "all", ["R2", "AIC", "RMSE"] Evaluation
plots list None ["forest_plot", "residuals_vs_fitted", "qq_plot"] Evaluation
vif_threshold float 5.0 Multicollinearity alert threshold Evaluation

Main Attributes (post-fit)

Attribute Content
model.summary_df Coefficients, CI 95%, p-values
model.performance_metrics_df R², AIC, BIC, RMSE, MSE, MAE
model.vif_df Variance Inflation Factor per predictor
model.cv_df Cross-validation metrics
model.assumption_metrics_df Durbin-Watson, Shapiro-Wilk, influential points

Common Metrics

Category Metric Attribute
Fit R² / Adjusted R² model.r2 / model.adjusted_r2
Fit AIC / BIC model.aic / model.bic
Error RMSE model.rmse
Error MAE model.mae
Assumption Durbin-Watson model.dw
Assumption Shapiro-Wilk p-value model.shapiro_wilk_p_value
Multicollinearity VIF model.vif_df
Family Link Use Case
gaussian identity Standard linear regression
gamma log Positively skewed, positive outcomes
tweedie log Mixed zeros + positive continuous

Minimal Example

from isaric.pipelines.factory import RAPID_PipelineFactory

factory = RAPID_PipelineFactory()

# Create model
model = factory.create(
    "glm",
    data=df,
    dependent_var="outcome",
    independent_vars=["age", "bmi", "sex"],
    family="gaussian",
    link="identity"
)

# Train
model.fit(labels={"age": "Age", "bmi": "BMI"})

# View results
model.summary(
    assumptions=["VIF", "Durbin-Watson"],
    performance=["R2", "AIC", "RMSE"],
    plots=["forest_plot"]
)

# Access coefficients
print(model.summary_df)
Want to... Go to...
Understand the theory? GLM Tutorial