Handle observableTransformation in petab.v1.simulate.sample_noise (#383)

Previously, `petab.v1.simulate.sample_noise` silently ignored `observableTransformation`.
For example, in case of observableTransformation=log and noiseDistribution=normal, it incorrectly sampled from a normal instead of a log-normal distribution.

Fixes #382.

---------

Co-authored-by: Dilan Pathirana <59329744+dilpath@users.noreply.github.com>

Author: dweindl

petab/v1/simulate.py

@@ -241,20 +241,39 @@ def sample_noise(
         simulated_value,
     )
 
-    # default noise distribution is petab.C.NORMAL
-    noise_distribution = petab_problem.observable_df.loc[
+    observable_row = petab_problem.observable_df.loc[
         measurement_row[petab.C.OBSERVABLE_ID]
-    ].get(petab.C.NOISE_DISTRIBUTION, petab.C.NORMAL)
+    ]
+    # default noise distribution is petab.C.NORMAL
+    noise_distribution = observable_row.get(
+        petab.C.NOISE_DISTRIBUTION, petab.C.NORMAL
+    )
     # an empty noise distribution column in an observables table can result in
     # `noise_distribution == float('nan')`
     if pd.isna(noise_distribution):
         noise_distribution = petab.C.NORMAL
 
+    observable_transformation = observable_row.get(
+        petab.C.OBSERVABLE_TRANSFORMATION, petab.C.LIN
+    )
+    transform = lambda x: x  # noqa: E731
+    # observableTransformation=log -> the log of the simulated value is
+    #  distributed according to `noise_distribution`
+    if observable_transformation == petab.C.LOG:
+        simulated_value = np.log(simulated_value)
+        transform = np.exp
+    elif observable_transformation == petab.C.LOG10:
+        simulated_value = np.log10(simulated_value)
+        transform = lambda x: np.power(10, x)  # noqa: E731
+
     # below is e.g.: `np.random.normal(loc=simulation, scale=noise_value)`
     simulated_value_with_noise = getattr(rng, noise_distribution)(
         loc=simulated_value, scale=noise_value * noise_scaling_factor
     )
 
+    # apply observable transformation, ensure `float` type
+    simulated_value_with_noise = float(transform(simulated_value_with_noise))
+
     if zero_bounded and np.sign(simulated_value) != np.sign(
         simulated_value_with_noise
     ):