API reference

Signatures and public methods are generated from this checkout. Start with the method selection guide for assumptions, data shapes and inference limits. Inherited methods may be unimplemented for a particular estimator; a signature alone is not a claim of support. Edit Python docstrings or the API inventory, not these generated pages.

Regression and estimation interfaces

Least squares, likelihood and fixed-effect operations.

BaseEstimator	Base class for all estimators in trex.
LinearRegression	Least squares with optional additive fixed effects.
MaximumLikelihoodEstimator	Base class for Maximum Likelihood Estimators using PyTorch optimizers.
LogisticRegression	Logistic Regression for binary classification with MLE estimation.
PoissonRegression	Poisson Regression for count data with MLE estimation.
demean_torch	Return weighted within residuals and a convergence flag.
prepare_fixed_effects	Encode each factor independently, including string or sparse numeric IDs.
demean.fixed_effect_rank	Exact rank of additive FE incidence, without dense n-by-level dummies.

Unconditional moments

GMM backends, GEL and their criterion functions.

gmm.GMMEstimator	Dispatch to a SciPy or Torch GMM implementation.
gmm.gmm.GMMEstimatorScipy
gmm.gmm.GMMEstimatorTorch
gmm.gmm.moment_covariance	Uncentered long-run covariance of observation-level moment vectors.
gmm.GELEstimator	Generalized empirical likelihood estimator for vector-valued moments.
gmm.rho_el	Empirical Likelihood (EL): rho(v) = log(1-v)
gmm.rho_exponential	Exponential tilting (ET): rho(v) = 1 - exp(v)
gmm.rho_cue	Continuously Updated Estimator (CUE): rho(v) = -0.5*v^2 - v
gmm.ScalarIVCUE	Scalar CUE objective `(a-betab)' Omega(beta)^-1 (a-betab)`.
gmm.PolynomialCUEResult	Candidate enumeration result; `beta=inf` means no selected finite fit.

Conditional moments and scores

MaximumMomentRestriction	Kernel maximum moment restriction estimator.
SieveMinimumDistance	Polynomial-sieve minimum distance estimator for conditional moments.
cmr.ConditionalMomentEstimator	Base class for neural conditional moment estimators.
cmr.rbf_kernel	RBF kernel matrix and bandwidth.
cmr.mmr_loss	Maximum moment restriction loss.
cmr.hsic	Biased HSIC statistic using RBF kernels.
cmr.polynomial_sieve	Polynomial basis expansion for instruments.
cmr.median_bandwidth	Median-heuristic bandwidth for an RBF kernel.
cmr.squared_distance	Pairwise squared Euclidean distance.
AntitonicScore	Piecewise-linear antitonic score estimate.
AntitonicScoreMatchingRegression	Linear regression via antitonic score matching.
estimate_antitonic_score	Estimate the antitonic projection of an error score from residuals.

Static and dynamic choice

choice.base.ChoiceModel	Base class for structural discrete choice models.
choice.BinaryLogit	Binary Logit model for discrete choice analysis.
choice.BinaryProbit	Binary Probit model for discrete choice analysis.
choice.MultinomialLogit	Multinomial Logit (MNL) model for discrete choice with multiple alternatives.
choice.LowRankLogit	Low-Rank Logit model for matrix completion style choice problems.
choice.DynamicChoiceData	Container for dynamic choice panel data.
choice.DynamicChoiceModel	Base class for dynamic discrete choice models.
choice.RustNFP	Rust (1987) full maximum likelihood via nested fixed point (NFP).
choice.HotzMillerCCP	Hotz & Miller (1993) CCP inversion estimator.
choice.LinearFlowUtility	Linear-in-features flow utility specification.
choice.ReplacementUtility	Rust-style bus replacement flow utility.
choice.estimate_transition_matrix	Estimate state transition probabilities P(x’\|x,a) from panel data.
choice.discretize_state	Discretize continuous state variable into bins.
choice.DeepValueFunction	Neural network value function approximation for high-dimensional states.
choice.estimate_ccps	Frequency CCPs on the input device; unobserved states use uniform rows.

Panel estimators and latent heterogeneity

NuclearNormMatrixCompletion	Nuclear-norm penalized matrix completion with additive fixed effects.
panel.MatrixCompletionResult
SyntheticDID	Synthetic difference-in-differences with two-way balancing weights.
panel.SyntheticDIDResult
did_estimate	Classical two-way DID estimate for a block treatment design.
sc_estimate	Synthetic-control special case: no time weights and no omega intercept.
matrix_completion_estimate	Estimate ATT by imputing treated post cells with matrix completion.
synthdid_estimate	Convenience function returning a fitted :class:`SyntheticDIDResult`.
panel_estimates	Compute the point-estimate row from the synthdid README estimator table.
panel.matrix_completion_lambda_max	Largest lambda for which the SVT update returns a zero low-rank matrix.
panel.simplex_least_squares_fw	Minimize \|\|Ax-b\|\|^2 + zeta^2 n \|\|x\|\|^2 over simplex by Frank-Wolfe.
panel.collapsed_form	Collapse treated units and post-treatment periods as in synthdid.
panel.frank_wolfe_step	One exact-line-search Frank-Wolfe step over the probability simplex.
LatentFactorGLM	Latent-factor GLM with factorized optimization.
KNNGroupedFixedEffects	Experimental grouped fixed-effects estimator with KNN-smoothed embeddings.
grouped_fe.GroupedFEResult	Container for grouped FE output.
build_panel_embeddings	Build unit-level embeddings by stacking panel moments over time.
grouped_fe.chunked_knn_indices	Compute exact KNN indices and squared distances in chunks.
grouped_fe.smooth_embeddings_by_knn	Average each embedding with its nearest neighbors.
grouped_fe.cluster_embeddings	Cluster embeddings with flash-kmeans when available, otherwise torch fallback.

Simulation, preprocessing and diagnostics

TabularTransformer	Standardize continuous columns while preserving binary columns.
TabularWGAN	Wasserstein GAN with gradient penalty for tabular simulation.
TabularPTGAN	Parallelly tempered Wasserstein GAN for multimodal tabular data.
TabularDiffusion	Small DDPM-style diffusion model for continuous tabular rows.
distribution_metrics	Compute marginal and joint distribution discrepancy metrics.
sliced_wasserstein_distance	Monte Carlo directional average of exact empirical 1-Wasserstein.
SafetensorsLLMInContextGenerator	Generate CSV-like tabular rows from a local safetensors language model.
SafetensorsQLORAGenerator	QLoRA-fine-tuned LLM row generator.