Validator: check for positive bounds for log-scaled parameter (#278)

Check the parameter table for positive bounds for log-scaled estimated
parameters that don't have an explicit intialization prior or that have initialPriorType=parameterScaleUniform.

See discussion in #259

Supersedes and closes #259

Author: dweindl

petab/lint.py

@@ -558,6 +558,17 @@ def check_parameter_bounds(parameter_df: pd.DataFrame) -> None:
                     f"Bounds for {row[PARAMETER_SCALE]} scaled parameter "
                     f"{ row.name} must be positive."
                 )
+            if (
+                row.get(PARAMETER_SCALE, LIN) in [LOG, LOG10]
+                and (row[LOWER_BOUND] == 0.0 or row[UPPER_BOUND] == 0.0)
+                and not row.get(INITIALIZATION_PRIOR_TYPE)
+            ):
+                raise AssertionError(
+                    f"Bounds for {row[PARAMETER_SCALE]} scaled parameter "
+                    f"{row.name} must be positive if no "
+                    f"{INITIALIZATION_PRIOR_TYPE} is provided. "
+                    "Cannot sample from unbounded interval."
+                )
 
 
 def assert_parameter_prior_type_is_valid(parameter_df: pd.DataFrame) -> None:
@@ -636,6 +647,19 @@ def assert_parameter_prior_parameters_are_valid(
                     f"{PARAMETER_SEPARATOR}par2' for all prior types)."
                 )
 
+            # we can't sample uniformly from [log(0)=-inf, ...]
+            if (
+                type_col == INITIALIZATION_PRIOR_TYPE
+                and row.get(type_col, "") == PARAMETER_SCALE_UNIFORM
+                and row.get(PARAMETER_SCALE, LIN) in [LOG, LOG10]
+                and (pars[0] == 0.0 or pars[1] == 0.0)
+            ):
+                raise AssertionError(
+                    f"{prior_par_cols} for {row[PARAMETER_SCALE]} scaled "
+                    f"parameter {row.name} must be positive if "
+                    f"{type_col}={PARAMETER_SCALE_UNIFORM}."
+                )
+
 
 def assert_parameter_estimate_is_boolean(parameter_df: pd.DataFrame) -> None:
     """

tests/test_lint.py

@@ -642,3 +642,18 @@ def test_parameter_ids_are_unique():
     parameter_df.index = ["par0", "par1"]
     parameter_df.index.name = "parameterId"
     lint.check_parameter_df(parameter_df)
+
+
+def test_check_positive_bounds_for_scaled_parameters():
+    parameter_df = pd.DataFrame(
+        {
+            PARAMETER_ID: ["par"],
+            PARAMETER_SCALE: [LOG10],
+            ESTIMATE: [1],
+            LOWER_BOUND: [0.0],
+            UPPER_BOUND: [1],
+        }
+    ).set_index(PARAMETER_ID)
+
+    with pytest.raises(AssertionError, match="positive"):
+        lint.check_parameter_df(parameter_df)