MaximumLikelihoodEstimator

Guide & worked example · Source

MaximumLikelihoodEstimator(
    optimizer=None,
    maxiter=5000,
    tol=0.0001,
    device=None,
)

Base class for Maximum Likelihood Estimators using PyTorch optimizers.

This class provides a flexible framework for fitting statistical models via maximum likelihood estimation. It supports various PyTorch optimizers and automatically computes standard errors using the Fisher information matrix.

Attributes: optimizer_class: PyTorch optimizer class (default: LBFGS). maxiter: Maximum number of optimization iterations. tol: Convergence tolerance for relative change in loss. params: Dictionary containing fitted parameters and diagnostics. history: Dictionary tracking optimization history (e.g., loss values).

Examples: >>> class MyModel(MaximumLikelihoodEstimator): … def _negative_log_likelihood(self, params, X, y): … return torch.sum((y - X @ params) ** 2) # OLS example … def _compute_fisher_information(self, params, X, y): … return X.T @ X >>> model = MyModel() >>> model.fit(X, y)

Attributes

Name	Description
device
history
maxiter
optimizer_class
params
tol

Methods

Name	Description
fit	Fit the model using maximum likelihood estimation.
summary	Print a formatted summary of model estimation results.
to	Move model parameters and fitted data to device.

fit

MaximumLikelihoodEstimator.fit(
    X,
    y,
    init_params=None,
    verbose=False,
    fe=None,
    fe_design=None,
    hdfe_index=None,
    offset=None,
    batch_size=None,
    stderr=True,
)

Fit the model using maximum likelihood estimation.

Optimizes the model parameters to minimize the negative log-likelihood using the specified PyTorch optimizer. After convergence, computes standard errors via the Fisher information matrix.

Args: X: Design matrix of shape (n_samples, n_features). Assumes X includes an intercept column if desired. y: Target vector of shape (n_samples,). init_params: Initial parameter values. For FE-aware GLMs, may be a packed parameter vector or a dict with coef and fe_coef. verbose: If True, prints convergence information. fe: Optional list of 1D fixed-effect ID vectors. fe_design: Optional list of sparse fixed-effect incidence matrices. hdfe_index: Optional index of the FE block to use for fast inference. offset: Optional offset vector added to the linear predictor. batch_size: Optional mini-batch size for first-order optimizers. stderr: Whether to compute standard errors.

Returns: Self (fitted estimator) with populated params dictionary.

Examples: >>> model = LogisticRegression() >>> model.fit(X_train, y_train, verbose=True) >>> print(model.params[“coef”])

summary

MaximumLikelihoodEstimator.summary(alpha=0.05)

Print a formatted summary of model estimation results.

Displays a comprehensive table including parameter estimates, standard errors, t-statistics, p-values, and confidence intervals (if standard errors are available).

Args: alpha: Significance level for confidence intervals (default: 0.05 for 95% confidence intervals).

Examples: >>> model.fit(X, y) >>> model.summary(alpha=0.01) # 99% confidence intervals

to

MaximumLikelihoodEstimator.to(device)

Move model parameters and fitted data to device.