fix yearsets.sample_from_poisson

CHANGELOG.md

@@ -23,6 +23,7 @@ Code freeze date: YYYY-MM-DD
 
 - `Hazard.from_xarray_raster` now stores strings as default values for `Hazard.event_name` [#795](https://github.com/CLIMADA-project/climada_python/pull/795)
 - Fix the dist_approx util function when used with method="geosphere" and log=True and points that are very close. [#792](https://github.com/CLIMADA-project/climada_python/pull/792)
+- `climada.util.yearsets.sample_from_poisson`: fix a bug ([#819](https://github.com/CLIMADA-project/climada_python/issues/819)) and inconsistency that occurs when lambda events per year (`lam`) are set to 1. [[#823](https://github.com/CLIMADA-project/climada_python/pull/823)]
 
 ### Deprecated
 

climada/util/test/test_yearsets.py

@@ -60,11 +60,14 @@ def test_impact_yearset_sampling_vect(self):
     def test_sample_from_poisson(self):
         """Test sampling amount of events per year."""
         n_sample_years = 1000
-        lam = np.sum(IMP.frequency)
-        events_per_year = yearsets.sample_from_poisson(n_sample_years, lam)
-
-        self.assertEqual(events_per_year.size, n_sample_years)
-        self.assertAlmostEqual(np.round(np.mean(events_per_year)), 2)
+        for lam in [0, 1, 2.5]:
+            events_per_year = yearsets.sample_from_poisson(n_sample_years, lam, seed=1)
+
+            self.assertEqual(events_per_year.size, n_sample_years)
+            self.assertAlmostEqual(np.mean(events_per_year), lam, places=1)
+
+        self.assertRaises(TypeError, yearsets.sample_from_poisson, n_sample_years, None)
+        self.assertRaises(ValueError, yearsets.sample_from_poisson, n_sample_years, -1)
 
     def test_sample_events(self):
         """Test the sampling of 34 events out of a pool of 20 events."""

climada/util/yearsets.py

@@ -152,7 +152,7 @@ def sample_from_poisson(n_sampled_years, lam, seed=None):
     -----------
         n_sampled_years : int
             The target number of years the impact yearset shall contain.
-        lam: int
+        lam: float
             the applied Poisson distribution is centered around lambda events per year
         seed : int, optional
             seed for numpy.random, will be set if not None
@@ -165,14 +165,8 @@ def sample_from_poisson(n_sampled_years, lam, seed=None):
     """
     if seed is not None:
         np.random.seed(seed)
-    if lam != 1:
-        events_per_year = np.round(np.random.poisson(lam=lam,
-                                                     size=n_sampled_years)).astype('int')
-    else:
-        events_per_year = np.ones(len(n_sampled_years))
-
+    return np.round(np.random.poisson(lam=lam, size=n_sampled_years)).astype('int')
 
-    return events_per_year
 
 def sample_events(events_per_year, freqs_orig, seed=None):
     """Sample events uniformely from an array (indices_orig) without replacement