marginaleffects 0.5.0__tar.gz → 0.5.1__tar.gz

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: marginaleffects
-Version: 0.5.0
+Version: 0.5.1
 Summary: Predictions, counterfactual comparisons, slopes, and hypothesis tests for statistical models.
 License-Expression: GPL-3.0-or-later
 Requires-Python: >=3.10

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/by.py

@@ -1,5 +1,4 @@
 import polars as pl
-import numpy as np
 from typing import List, Optional, Tuple
 
 
@@ -66,8 +65,7 @@ def _get_by_internal(
     else:
         out = pl.DataFrame({"estimate": estimand["estimate"]})
 
-    by = [x for x in by if x in out.columns]
-    by = np.unique(by)
+    by = list(dict.fromkeys(x for x in by if x in out.columns))
 
     if isinstance(by, list) and len(by) == 0:
         if return_groups and "rowid" in out.columns:

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/comparisons.py

@@ -1,4 +1,5 @@
 import re
+import warnings
 from functools import reduce
 
 import numpy as np
@@ -118,28 +119,22 @@ def _build_comparison_frames(newdata, variables, cross):
                 hi.append(hi_row)
                 lo.append(lo_row)
         else:
-            hi.append(newdata)
-            lo.append(newdata)
-            nd.append(newdata)
+            nd_row = newdata.clone()
+            hi_row = newdata.clone()
+            lo_row = newdata.clone()
             for v in variables:
                 vcomp = "custom" if callable(v.comparison) else v.comparison
-                nd[0] = nd[0].with_columns(
+                shared = [
                     pl.lit(v.variable).alias("term"),
                     pl.lit(v.lab).alias(f"contrast_{v.variable}"),
                     pl.lit(vcomp).alias("marginaleffects_comparison"),
-                )
-                hi[0] = hi[0].with_columns(
-                    pl.lit(v.hi).alias(v.variable),
-                    pl.lit(v.variable).alias("term"),
-                    pl.lit(v.lab).alias(f"contrast_{v.variable}"),
-                    pl.lit(vcomp).alias("marginaleffects_comparison"),
-                )
-                lo[0] = lo[0].with_columns(
-                    pl.lit(v.lo).alias(v.variable),
-                    pl.lit(v.variable).alias("term"),
-                    pl.lit(v.lab).alias(f"contrast_{v.variable}"),
-                    pl.lit(vcomp).alias("marginaleffects_comparison"),
-                )
+                ]
+                nd_row = nd_row.with_columns(*shared)
+                hi_row = hi_row.with_columns(pl.lit(v.hi).alias(v.variable), *shared)
+                lo_row = lo_row.with_columns(pl.lit(v.lo).alias(v.variable), *shared)
+            nd.append(nd_row)
+            hi.append(hi_row)
+            lo.append(lo_row)
     return nd, hi, lo
 
 
@@ -166,9 +161,10 @@ def _finalize_counterfactual_frames(
     pad_df = upcast(pad_df, hi)
     nd = upcast(nd, hi)
 
-    dfs_to_align = [("nd", nd), ("hi", hi), ("lo", lo)]
+    dfs = {"nd": nd, "hi": hi, "lo": lo}
 
-    for df_name, df in dfs_to_align:
+    for df_name in dfs:
+        df = dfs[df_name]
         common_cols = set(pad_df.columns) & set(df.columns)
         for col in common_cols:
             pad_dtype = str(pad_df[col].dtype)
@@ -189,8 +185,8 @@ def _finalize_counterfactual_frames(
                                 .alias(col)
                             )
                     except Exception as e:
-                        print(
-                            f"Warning: Could not convert List column {col} to strings: {e}"
+                        warnings.warn(
+                            f"Could not convert List column {col} to strings: {e}"
                         )
                         try:
                             if col in pad_df.columns and pad_df.height > 0:
@@ -198,7 +194,7 @@ def _finalize_counterfactual_frames(
                             if col in df.columns and df.height > 0:
                                 df = df.explode(col)
                         except Exception as e2:
-                            print(f"Warning: Could not explode List column {col}: {e2}")
+                            warnings.warn(f"Could not explode List column {col}: {e2}")
                             if col in pad_df.columns:
                                 pad_df = pad_df.with_columns(
                                     pad_df[col].cast(pl.String).alias(col)
@@ -206,12 +202,9 @@ def _finalize_counterfactual_frames(
                             if col in df.columns:
                                 df = df.with_columns(df[col].cast(pl.String).alias(col))
 
-        if df_name == "nd":
-            nd = df
-        elif df_name == "hi":
-            hi = df
-        elif df_name == "lo":
-            lo = df
+        dfs[df_name] = df
+
+    nd, hi, lo = dfs["nd"], dfs["hi"], dfs["lo"]
 
     nd = pl.concat([pad_df, nd], how="diagonal")
     hi = pl.concat([pad_df, hi], how="diagonal")
@@ -221,9 +214,7 @@ def _finalize_counterfactual_frames(
     categorical_list_cols = []
     for col in list_cols:
         dtype_str = str(nd[col].dtype)
-        if (
-            "Enum(" in dtype_str or "String" in dtype_str or "UInt32" in dtype_str
-        ) and col in ["Region"]:
+        if "Enum(" in dtype_str or "String" in dtype_str or "UInt32" in dtype_str:
             categorical_list_cols.append(col)
 
     if categorical_list_cols:
@@ -241,7 +232,7 @@ def _prepare_design_matrices(model, nd, hi, lo, pad_rows):
     lo_X = model.get_exog(lo)
     nd_X = model.get_exog(nd)
 
-    if pad_rows >= 0:
+    if pad_rows > 0:
         nd_X = nd_X[pad_rows:]
         hi_X = hi_X[pad_rows:]
         lo_X = lo_X[pad_rows:]

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/estimands.py

@@ -42,12 +42,12 @@ estimands = {
     "ratio": lambda hi, lo, eps, x, y, w: prep(hi / lo),
     "ratioavg": lambda hi, lo, eps, x, y, w: prep(hi.mean() / lo.mean()),
     "ratioavgwts": lambda hi, lo, eps, x, y, w: prep(
-        (hi * w).sum() / w.sum() / (lo * w).sum() / w.sum()
+        ((hi * w).sum() / w.sum()) / ((lo * w).sum() / w.sum())
     ),
     "lnratio": lambda hi, lo, eps, x, y, w: prep(np.log(hi / lo)),
     "lnratioavg": lambda hi, lo, eps, x, y, w: prep(np.log(hi.mean() / lo.mean())),
     "lnratioavgwts": lambda hi, lo, eps, x, y, w: prep(
-        np.log((hi * w).sum() / w.sum() / (lo * w).sum() / w.sum())
+        np.log(((hi * w).sum() / w.sum()) / ((lo * w).sum() / w.sum()))
     ),
     "lnor": lambda hi, lo, eps, x, y, w: prep(
         np.log((hi / (1 - hi)) / (lo / (1 - lo)))
@@ -69,7 +69,7 @@ estimands = {
     "expdydxavg": lambda hi, lo, eps, x, y, w: prep(
         np.mean(((hi.exp() - lo.exp()) / np.exp(eps)) / eps)
     ),
-    "expdydxavgwts": lambda hi, lo, eps, x, y, w: (
-        prep((((np.exp(hi) - np.exp(lo)) / np.exp(eps)) / eps) * w).sum() / w.sum()
+    "expdydxavgwts": lambda hi, lo, eps, x, y, w: prep(
+        ((((np.exp(hi) - np.exp(lo)) / np.exp(eps)) / eps) * w).sum() / w.sum()
     ),
 }

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/plot/common.py

@@ -1,3 +1,5 @@
+import copy
+
 import numpy as np
 from ..datagrid import datagrid  # noqa
 from ..sanitize import sanitize_model
@@ -7,7 +9,7 @@ import polars as pl
 def dt_on_condition(model, condition):
     model = sanitize_model(model)
 
-    condition_new = condition  # two pointers to the same object? this looks like a bug
+    condition_new = copy.deepcopy(condition)
 
     # not sure why newdata gets added
     modeldata = model.get_modeldata()
@@ -19,28 +21,24 @@ def dt_on_condition(model, condition):
     first_key = ""  # special case when the first element is numeric
 
     if isinstance(condition_new, list):
-        assert all(ele in modeldata.columns for ele in condition_new), (
-            "All elements of condition must be columns of the model."
-        )
+        if not all(ele in modeldata.columns for ele in condition_new):
+            raise ValueError("All elements of condition must be columns of the model.")
         first_key = condition_new[0]
         to_datagrid = {key: None for key in condition_new}
 
     elif isinstance(condition_new, dict):
-        assert all(key in modeldata.columns for key in condition_new.keys()), (
-            "All keys of condition must be columns of the model."
-        )
+        if not all(key in modeldata.columns for key in condition_new.keys()):
+            raise ValueError("All keys of condition must be columns of the model.")
         first_key = next(iter(condition_new))
-        to_datagrid = (
-            condition_new  # third pointer to the same object? looks like a BUG
-        )
+        to_datagrid = condition_new
 
-    # not sure why `newdata` sometimes gets added
     if isinstance(condition_new, dict) and "newdata" in to_datagrid.keys():
         condition_new.pop("newdata", None)
 
-    assert 1 <= len(condition_new) <= 4, (
-        f"Lenght of condition must be inclusively between 1 and 4. Got : {len(condition_new)}."
-    )
+    if not (1 <= len(condition_new) <= 4):
+        raise ValueError(
+            f"Length of condition must be inclusively between 1 and 4. Got: {len(condition_new)}."
+        )
 
     for key, value in to_datagrid.items():
         variable_type = model.get_variable_type(key)
@@ -51,20 +49,17 @@ def dt_on_condition(model, condition):
             )
 
         elif variable_type in ["character"]:
-            # get specified names of the condition
-            # here is the BUG, we take the values of "species" back from the model
             to_datagrid[key] = (
                 to_datagrid[key]
                 if to_datagrid[key]
                 else modeldata[key].unique().sort().to_list()
             )
-            assert len(to_datagrid[key]) <= 10, (
-                f"Character type variables of more than 10 unique values are not supported. {key} variable has {len(to_datagrid[key])} unique values."
-            )
+            if len(to_datagrid[key]) > 10:
+                raise ValueError(
+                    f"Character type variables of more than 10 unique values are not supported. {key} variable has {len(to_datagrid[key])} unique values."
+                )
 
         elif variable_type in ["boolean", "binary"]:
-            # get specified names of the condition
-            # here is the BUG, we take the values of "species" back from the model
             if to_datagrid[key] is None:
                 to_datagrid[key] = modeldata[key].unique().sort().to_list()
 
@@ -131,15 +126,14 @@ def ordered_cat(dt, k, lab):
 
 
 def validate_plot_args(condition, by, newdata, wts):
-    assert not (not by and newdata is not None), (
-        "The `newdata` argument requires a `by` argument."
-    )
-    assert not (wts is not None and not by), (
-        "The `wts` argument requires a `by` argument."
-    )
-    assert (condition is None and by) or (condition is not None and not by), (
-        "One of the `condition` and `by` arguments must be supplied, but not both."
-    )
+    if not by and newdata is not None:
+        raise ValueError("The `newdata` argument requires a `by` argument.")
+    if wts is not None and not by:
+        raise ValueError("The `wts` argument requires a `by` argument.")
+    if not ((condition is None and by) or (condition is not None and not by)):
+        raise ValueError(
+            "One of the `condition` and `by` arguments must be supplied, but not both."
+        )
 
 
 def extract_var_list(condition, by):
@@ -158,9 +152,10 @@ def extract_var_list(condition, by):
 
     var_list = [x for x in var_list if x not in ["newdata", "model"]]
 
-    assert len(var_list) < 5, (
-        "The `condition` and `by` arguments can have a max length of 4."
-    )
+    if len(var_list) >= 5:
+        raise ValueError(
+            "The `condition` and `by` arguments can have a max length of 4."
+        )
 
     return var_list
 
@@ -286,10 +281,10 @@ def plot_common(model, dt, y_label, var_list, gray=False, points=0):
         if len(var_list) > 1:
             if gray:
                 # get the number of unique values in the column "var_list[1]"
-                unique_values = dt[var_list[1]].unique().len()
-                if unique_values > 5:
+                unique_values = dt[var_list[1]].unique()
+                if unique_values.len() > 5:
                     raise ValueError(
-                        f"The number of elements in the second position of the `condition` or `by` argument (variable {var_list[1]}) cannot exceed 5. It has currently {len(unique_values)} elements, with values {unique_values}."
+                        f"The number of elements in the second position of the `condition` or `by` argument (variable {var_list[1]}) cannot exceed 5. It has currently {unique_values.len()} elements, with values {unique_values.to_list()}."
                     )
                 custom_line_types = [
                     "solid",

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/sanitize/comparison.py

@@ -73,8 +73,7 @@ def sanitize_comparison(comparison, by, wts=None):
         "expdydx": "exp(dY/dX)",
     }
 
-    assert out in lab.keys(), (
-        f"`comparison` must be one of: {', '.join(list(lab.keys()))}."
-    )
+    if out not in lab.keys():
+        raise ValueError(f"`comparison` must be one of: {', '.join(list(lab.keys()))}.")
 
     return (out, lab[out])

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/sanitize/newdata.py

@@ -1,12 +1,16 @@
 import numpy as np
 import polars as pl
 
-from ..datagrid import datagrid
-from ..utils import ingest, upcast
 from ..formula import listwise_deletion
 
 
 def sanitize_newdata(model, newdata, wts, by=[]):
+    # Lazy imports to break the `datagrid -> utils -> sanitize -> newdata -> ...`
+    # circular import that fires when `datagrid` is the first symbol pulled from
+    # marginaleffects in a fresh interpreter (see GH #1724).
+    from ..datagrid import datagrid
+    from ..utils import ingest, upcast
+
     modeldata = model.get_modeldata()
 
     if newdata is None:
@@ -72,9 +76,10 @@ def sanitize_newdata(model, newdata, wts, by=[]):
         "contrast",
         "statistic",
     }
-    assert not (set(out.columns) & reserved_names), (
-        f"Input data contain reserved column name(s) : {set(out.columns).intersection(reserved_names)}"
-    )
+    if set(out.columns) & reserved_names:
+        raise ValueError(
+            f"Input data contain reserved column name(s): {set(out.columns).intersection(reserved_names)}"
+        )
 
     datagrid_explicit = None
     if isinstance(newdata, pl.DataFrame) and hasattr(newdata, "datagrid_explicit"):

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/sanitize/variables.py

@@ -11,17 +11,11 @@ HiLo = namedtuple("HiLo", ["variable", "hi", "lo", "lab", "pad", "comparison"])
 
 
 def _clean_global(k, n):
-    if (
-        not isinstance(k, list)
-        and not isinstance(k, pl.Series)
-        and not isinstance(k, np.ndarray)
-    ):
-        out = [k]
+    if isinstance(k, (pl.Series, np.ndarray)):
+        return pl.Series(k) if len(k) > 1 else pl.Series(np.repeat(k[0], n))
     if not isinstance(k, list) or len(k) == 1:
-        out = pl.Series(np.repeat(k, n))
-    else:
-        out = pl.Series(k)
-    return out
+        return pl.Series(np.repeat(k, n))
+    return pl.Series(k)
 
 
 def _get_one_variable_hi_lo(
@@ -153,9 +147,10 @@ def _get_one_variable_hi_lo(
 
         elif callable(value):
             tmp = value(newdata[variable])
-            assert tmp.shape[1] == 2, (
-                f"The function passed to `variables` must return a DataFrame with two columns. Got {tmp.shape[1]}."
-            )
+            if tmp.shape[1] != 2:
+                raise ValueError(
+                    f"The function passed to `variables` must return a DataFrame with two columns. Got {tmp.shape[1]}."
+                )
             lo = tmp[:, 0]
             hi = tmp[:, 1]
             lab = "custom"
@@ -225,9 +220,8 @@ def sanitize_variables(
             )
 
     elif isinstance(variables, dict):
-        for v in variables:
+        for v in list(variables.keys()):
             if v not in newdata.columns:
-                del variables[v]
                 warn(f"Variable {v} is not in newdata.")
             else:
                 out.append(

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/sanitize/vcov.py

@@ -14,5 +14,6 @@ def sanitize_vcov(vcov, model):
 
     V = model.get_vcov(vcov)
     if V is not None:
-        assert isinstance(V, np.ndarray), "vcov must be True or a square NumPy array"
+        if not isinstance(V, np.ndarray):
+            raise TypeError("vcov must be True or a square NumPy array")
     return V

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/uncertainty.py

@@ -6,36 +6,29 @@ import scipy.stats as stats
 
 
 def get_jacobian(func, coefs, eps_vcov=None):
-    # forward finite difference (faster)
+    original_shape = None
     if coefs.ndim == 2:
+        original_shape = coefs.shape
         if isinstance(coefs, np.ndarray):
             coefs_flat = coefs.flatten(order="F")
         else:
             coefs_flat = coefs.to_numpy().flatten(order="F")
-        baseline = func(coefs)["estimate"].to_numpy()
-        jac = np.empty((baseline.shape[0], len(coefs_flat)), dtype=np.float64)
-        for i, xi in enumerate(coefs_flat):
-            if eps_vcov is not None:
-                h = eps_vcov
-            else:
-                h = max(abs(xi) * np.sqrt(np.finfo(float).eps), 1e-10)
-            dx = np.copy(coefs_flat)
-            dx[i] = dx[i] + h
-            tmp = dx.reshape(coefs.shape, order="F")
-            jac[:, i] = (func(tmp)["estimate"].to_numpy() - baseline) / h
-        return jac
     else:
-        baseline = func(coefs)["estimate"].to_numpy()
-        jac = np.empty((baseline.shape[0], len(coefs)), dtype=np.float64)
-        for i, xi in enumerate(coefs):
-            if eps_vcov is not None:
-                h = eps_vcov
-            else:
-                h = max(abs(xi) * np.sqrt(np.finfo(float).eps), 1e-10)
-            dx = np.copy(coefs)
-            dx[i] = dx[i] + h
-            jac[:, i] = (func(dx)["estimate"].to_numpy() - baseline) / h
-        return jac
+        coefs_flat = np.asarray(coefs)
+
+    baseline = func(coefs)["estimate"].to_numpy()
+    jac = np.empty((baseline.shape[0], len(coefs_flat)), dtype=np.float64)
+    for i, xi in enumerate(coefs_flat):
+        if eps_vcov is not None:
+            h = eps_vcov
+        else:
+            h = max(abs(xi) * np.sqrt(np.finfo(float).eps), 1e-10)
+        dx = np.copy(coefs_flat)
+        dx[i] = dx[i] + h
+        if original_shape is not None:
+            dx = dx.reshape(original_shape, order="F")
+        jac[:, i] = (func(dx)["estimate"].to_numpy() - baseline) / h
+    return jac
 
 
 def get_se(J, V):
@@ -65,7 +58,7 @@ def get_z_p_ci(df, model, conf_level, hypothesis_null=0):
             "statistic"
         )
     )
-    if hasattr(model, "df_resid") and isinstance(model.df_resid, float):
+    if hasattr(model, "df_resid") and isinstance(model.df_resid, (int, float)):
         dof = model.df_resid
     else:
         dof = np.inf
@@ -93,6 +86,6 @@ def get_z_p_ci(df, model, conf_level, hypothesis_null=0):
                 .map_batches(lambda x: -np.log2(x), return_dtype=pl.Float64)
                 .alias("s_value")
             )
-        except Exception as e:
-            print(f"An exception occurred: {e}")
+        except Exception:
+            pass
     return df

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects/utils.py

@@ -116,7 +116,7 @@ def upcast(df, reference):
         pl.Float64,
     ]
     for col in df.columns:
-        if col in df.columns and col in reference.columns:
+        if col in reference.columns:
             good = reference[col].dtype
             bad = df[col].dtype
             if good != bad:

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/marginaleffects.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: marginaleffects
-Version: 0.5.0
+Version: 0.5.1
 Summary: Predictions, counterfactual comparisons, slopes, and hypothesis tests for statistical models.
 License-Expression: GPL-3.0-or-later
 Requires-Python: >=3.10

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "marginaleffects"
-version = "0.5.0"
+version = "0.5.1"
 license = "GPL-3.0-or-later"
 description = "Predictions, counterfactual comparisons, slopes, and hypothesis tests for statistical models."
 readme = "README.md"

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/tests/test_bugfix.py

@@ -1,3 +1,6 @@
+import subprocess
+import sys
+
 import numpy as np
 import pandas as pd
 import polars as pl
@@ -22,3 +25,18 @@ def test_issue_226_np_context():
     out = predictions(mod, newdata=df)
     assert isinstance(out, MarginaleffectsResult)
     assert isinstance(out.data, pl.DataFrame)
+
+
+def test_issue_1724():
+    # Circular import when `datagrid` is the first symbol pulled from
+    # marginaleffects in a fresh interpreter. Must run in a subprocess —
+    # the in-process pytest run has already warmed the import graph.
+    result = subprocess.run(
+        [sys.executable, "-c", "from marginaleffects import datagrid"],
+        capture_output=True,
+        text=True,
+    )
+    assert result.returncode == 0, (
+        f"Fresh-process import of `datagrid` failed.\n"
+        f"stdout: {result.stdout}\nstderr: {result.stderr}"
+    )

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/tests/test_comparisons.py

@@ -200,7 +200,7 @@ def test_lift():
     cmp2 = comparisons(mod, comparison="liftavg")
     assert cmp1.shape[0] == 32
     assert cmp2.shape[0] == 1
-    with pytest.raises(AssertionError):
+    with pytest.raises(ValueError):
         comparisons(mod, comparison="liftr")
 
 

{marginaleffects-0.5.0 → marginaleffects-0.5.1}/tests/test_pyfixest.py

@@ -204,7 +204,7 @@ def test_pyfixest_standard_errors_across_models():
     fit_pois_fe = fepois("Y ~ X1 * X2 * Z1 | f1", data=poisson_data)
     with pytest.warns(
         UserWarning,
-        match="uncertainty in fixed-effects parameters when computing contrasts",
+        match="cannot take into account the uncertainty in fixed-effects",
     ):
         try:
             comp_pois_fe = comparisons(fit_pois_fe)