apply to v1

Author: dweindl

petab/v1/calculate.py

@@ -140,19 +140,17 @@ def calculate_residuals_for_table(
             # apply scaling
             observable = observable_df.loc[row[OBSERVABLE_ID]]
             trafo = observable.get(OBSERVABLE_TRANSFORMATION, LIN)
-            simulation = petab.scale(simulation, trafo)
-            measurement = petab.scale(measurement, trafo)
+            scaled_simulation = petab.scale(simulation, trafo)
+            scaled_measurement = petab.scale(measurement, trafo)
 
         # non-normalized residual is just the difference
-        residual = simulation - measurement
+        residual = scaled_measurement - scaled_simulation
 
-        noise_value = 1
         if normalize:
-            # look up noise standard deviation
-            noise_value = evaluate_noise_formula(
+            # divide by standard deviation
+            residual /= evaluate_noise_formula(
                 row, noise_formulas, parameter_df, simulation
             )
-        residual /= noise_value
 
         # fill in value
         residual_df.loc[irow, RESIDUAL] = residual
@@ -170,13 +168,10 @@ def get_symbolic_noise_formulas(observable_df) -> dict[str, sp.Expr]:
     """
     noise_formulas = {}
     # iterate over observables
-    for row in observable_df.itertuples():
-        observable_id = row.Index
-        if NOISE_FORMULA not in observable_df.columns:
-            noise_formula = None
-        else:
-            noise_formula = sympify_petab(row.noiseFormula)
-        noise_formulas[observable_id] = noise_formula
+    for observable_id, row in observable_df.iterrows():
+        noise_formulas[observable_id] = (
+            sympify_petab(row.noiseFormula) if NOISE_FORMULA in row else None
+        )
     return noise_formulas
 
 
@@ -376,7 +371,7 @@ def calculate_llh_for_table(
 
         # get noise standard deviation
         noise_value = evaluate_noise_formula(
-            row, noise_formulas, parameter_df, petab.scale(simulation, scale)
+            row, noise_formulas, parameter_df, simulation
         )
 
         # get noise distribution

tests/v1/test_calculate.py

@@ -43,12 +43,12 @@ def model_simple():
     simulation_df[SIMULATION] = [2, 2, 19, 20]
 
     expected_residuals = {
-        (2 - 0) / 2,
-        (2 - 1) / 2,
-        (19 - 20) / 3,
-        (20 - 22) / 3,
+        (0 - 2) / 2,
+        (1 - 2) / 2,
+        (20 - 19) / 3,
+        (22 - 20) / 3,
     }
-    expected_residuals_nonorm = {2 - 0, 2 - 1, 19 - 20, 20 - 22}
+    expected_residuals_nonorm = {0 - 2, 1 - 2, 20 - 19, 22 - 20}
     expected_llh = (
         -0.5 * (np.array(list(expected_residuals)) ** 2).sum()
         - 0.5 * np.log(2 * np.pi * np.array([2, 2, 3, 3]) ** 2).sum()
@@ -93,8 +93,8 @@ def model_replicates():
     )
     simulation_df[SIMULATION] = [2, 2]
 
-    expected_residuals = {(2 - 0) / 2, (2 - 1) / 2}
-    expected_residuals_nonorm = {2 - 0, 2 - 1}
+    expected_residuals = {(0 - 2) / 2, (1 - 2) / 2}
+    expected_residuals_nonorm = {0 - 2, 1 - 2}
     expected_llh = (
         -0.5 * (np.array(list(expected_residuals)) ** 2).sum()
         - 0.5 * np.log(2 * np.pi * np.array([2, 2]) ** 2).sum()
@@ -141,12 +141,12 @@ def model_scalings():
     simulation_df[SIMULATION] = [2, 3]
 
     expected_residuals = {
-        (np.log(2) - np.log(0.5)) / 2,
-        (np.log(3) - np.log(1)) / 2,
+        (np.log(0.5) - np.log(2)) / 2,
+        (np.log(1) - np.log(3)) / 2,
     }
     expected_residuals_nonorm = {
-        np.log(2) - np.log(0.5),
-        np.log(3) - np.log(1),
+        np.log(0.5) - np.log(2),
+        np.log(1) - np.log(3),
     }
     expected_llh = (
         -0.5 * (np.array(list(expected_residuals)) ** 2).sum()
@@ -201,21 +201,20 @@ def model_non_numeric_overrides():
     simulation_df[SIMULATION] = [2, 3]
 
     expected_residuals = {
-        (np.log(2) - np.log(0.5)) / (2 * 7 + 8 + 4 + np.log(2)),
-        (np.log(3) - np.log(1)) / (2 * 2 + 3 + 4 + np.log(3)),
+        (np.log(0.5) - np.log(2)) / (2 * 7 + 8 + 4 + 2),
+        (np.log(1) - np.log(3)) / (2 * 2 + 3 + 4 + 3),
     }
     expected_residuals_nonorm = {
-        np.log(2) - np.log(0.5),
-        np.log(3) - np.log(1),
+        np.log(0.5) - np.log(2),
+        np.log(1) - np.log(3),
     }
     expected_llh = (
         -0.5 * (np.array(list(expected_residuals)) ** 2).sum()
         - 0.5
         * np.log(
             2
             * np.pi
-            * np.array([2 * 7 + 8 + 4 + np.log(2), 2 * 2 + 3 + 4 + np.log(3)])
-            ** 2
+            * np.array([2 * 7 + 8 + 4 + 2, 2 * 2 + 3 + 4 + 3]) ** 2
             * np.array([0.5, 1]) ** 2
         ).sum()
     )
@@ -261,8 +260,8 @@ def model_custom_likelihood():
     )
     simulation_df[SIMULATION] = [2, 3]
 
-    expected_residuals = {(np.log(2) - np.log(0.5)) / 2, (3 - 2) / 1.5}
-    expected_residuals_nonorm = {np.log(2) - np.log(0.5), 3 - 2}
+    expected_residuals = {(np.log(0.5) - np.log(2)) / 2, (2 - 3) / 1.5}
+    expected_residuals_nonorm = {np.log(0.5) - np.log(2), 2 - 3}
     expected_llh = (
         -np.abs(list(expected_residuals)).sum()
         - np.log(2 * np.array([2, 1.5]) * np.array([0.5, 1])).sum()