Merge pull request #660 from leostimpfle/iterative-rectifier-2

Reopen iterative rectifier PR

pyfixest/estimation/FixestMulti_.py

@@ -28,6 +28,7 @@ def __init__(
         seed: Optional[int],
         split: Optional[str],
         fsplit: Optional[str],
+        separation_check: Optional[list[str]] = None,
     ) -> None:
         """
         Initialize a class for multiple fixed effect estimations.
@@ -57,6 +58,9 @@ def __init__(
         seed : Optional[int]
             Option to provide a random seed. Default is None.
             Only relevant for wild cluster bootstrap for use_compression=True.
+        separation_check: list[str], optional
+            Only used in "fepois". Methods to identify and drop separated observations.
+            Either "fe" or "ir". Executes both by default.
 
         Returns
         -------
@@ -70,6 +74,7 @@ def __init__(
         self._use_compression = use_compression
         self._reps = reps if use_compression else None
         self._seed = seed if use_compression else None
+        self._separation_check = separation_check
 
         self._run_split = split is not None or fsplit is not None
         self._run_full = not (split and not fsplit)
@@ -186,6 +191,7 @@ def _estimate_all_models(
         collin_tol: float = 1e-6,
         iwls_maxiter: int = 25,
         iwls_tol: float = 1e-08,
+        separation_check: Optional[list[str]] = None,
     ) -> None:
         """
         Estimate multiple regression models.
@@ -208,6 +214,9 @@ def _estimate_all_models(
         iwls_tol : float, optional
             The tolerance level for the IWLS algorithm. Default is 1e-8.
             Only relevant for non-linear estimation strategies.
+        separation_check: list[str], optional
+            Only used in "fepois". Methods to identify and drop separated observations.
+            Either "fe" or "ir". Executes both by default.
 
         Returns
         -------
@@ -318,6 +327,7 @@ def _estimate_all_models(
                             lean=_lean,
                             sample_split_value=sample_split_value,
                             sample_split_var=_splitvar,
+                            separation_check=separation_check,
                             # solver=_solver
                         )
                         FIT.prepare_model_matrix()

pyfixest/estimation/estimation.py

@@ -419,6 +419,7 @@ def fepois(
     iwls_tol: float = 1e-08,
     iwls_maxiter: int = 25,
     collin_tol: float = 1e-10,
+    separation_check: Optional[list[str]] = ["fe"],
     solver: str = "np.linalg.solve",
     drop_intercept: bool = False,
     i_ref1=None,
@@ -471,6 +472,10 @@ def fepois(
     collin_tol : float, optional
         Tolerance for collinearity check, by default 1e-10.
 
+    separation_check: list[str], optional
+        Methods to identify and drop separated observations.
+        Either "fe" or "ir". Executes "fe" by default.
+
     solver : str, optional.
         The solver to use for the regression. Can be either "np.linalg.solve" or
         "np.linalg.lstsq". Defaults to "np.linalg.solve".
@@ -569,6 +574,7 @@ def fepois(
         seed=None,
         split=split,
         fsplit=fsplit,
+        separation_check=separation_check,
     )
 
     fixest = FixestMulti(
@@ -598,6 +604,7 @@ def fepois(
         iwls_tol=iwls_tol,
         iwls_maxiter=iwls_maxiter,
         collin_tol=collin_tol,
+        separation_check=separation_check,
         solver=solver,
     )
 
@@ -625,6 +632,7 @@ def _estimation_input_checks(
     seed: Optional[int],
     split: Optional[str],
     fsplit: Optional[str],
+    separation_check: Optional[list[str]] = None,
 ):
     if not isinstance(fml, str):
         raise TypeError("fml must be a string")
@@ -726,3 +734,14 @@ def _estimation_input_checks(
 
     if isinstance(fsplit, str) and fsplit not in data.columns:
         raise KeyError(f"Column '{fsplit}' not found in data.")
+
+    if separation_check is not None:
+        if not isinstance(separation_check, list):
+            raise TypeError(
+                "The function argument `separation_check` must be of type list."
+            )
+
+        if not all(x in ["fe", "ir"] for x in separation_check):
+            raise ValueError(
+                "The function argument `separation_check` must be a list of strings containing 'fe' and/or 'ir'."
+            )