Fit the inverse-probability weight factor for a Parquet cohort in Rust and write the per-(id, period) factor table (id, period, weight_factor).

A thin FFI shim over tte_expand::fit_weights_parquet (the weights-fit surface). Unlike expand_weighted_parquet(), which applies a pre-computed factor table, this fits the IPW models in Rust: it ports TrialEmulation's data_manipulation + censor_func design preparation and binds a deterministic binomial-logit solver for the switching and/or IPCW censoring models, then forms wt = wt_switch * wtC. The structural design is exact; the fitted factors reproduce R glm within the staged ~1e-6 tolerance (ADR-2), not bit-for-bit. Robust/sandwich variance and the marginal structural model stay in R.

Usage

fit_weights_parquet(
  input_path,
  output_path,
  id_col,
  period_col,
  treatment_col,
  eligible_col,
  outcome_col,
  first_period,
  last_period,
  estimand,
  use_switch,
  switch_numerator,
  switch_denominator,
  use_censor,
  censor_col,
  censor_numerator,
  censor_denominator,
  pool_censor
)

Arguments

input_path

Path to the input Parquet cohort (long person-time).

output_path

Path where the (id, period, weight_factor) table is written.

id_col, period_col, treatment_col

Column names in the input.

eligible_col, outcome_col

Eligibility / outcome column names.

first_period, last_period

Inclusive integer bounds on trial_period.

estimand

"ITT" or "PP"; per-protocol runs the artificial-censoring state machine and (with switching covariates) the switch models. Case-insensitive.

use_switch

Whether to fit per-protocol switching-weight models.

switch_numerator, switch_denominator

Covariate columns for the switching numerator (stabiliser) and denominator models (ignored when use_switch is FALSE).

use_censor

Whether to fit inverse-probability-of-censoring (IPCW) models.

censor_col

Name of the {0,1} censoring-indicator column; the response is 1 - censor_col (ignored when use_censor is FALSE).

censor_numerator, censor_denominator

Covariate columns for the IPCW numerator/denominator models (ignored when use_censor is FALSE).

pool_censor

How the IPCW models are pooled across the previous-treatment strata: "none", "numerator", or "both". Case-insensitive.

Value

NULL, invisibly; the factor table is written to output_path. Errors in the core engine (including weight-fit failures) surface as R errors.

Examples

if (FALSE) { # \dontrun{
fit_weights_parquet(
  "cohort.parquet", "factors.parquet",
  "id", "period", "treatment", "eligible", "outcome",
  0L, .Machine$integer.max, "PP",
  TRUE, c("x2"), c("x2", "x1"),
  FALSE, "", character(0), character(0), "none"
)
} # }