Merge pull request #354 from LSSTDESC/issue/41/rando

Addressing the last few remaining places where randoms aren't using default_rng and/or a configurable seed

src/rail/creation/degradation/grid_selection.py

@@ -73,7 +73,7 @@ def run(self):
         If using a color-based redshift cut, galaxies with redshifts > the percentile cut are removed from the sample
         before making the random selection.
         """
-        np.random.seed(self.config.random_seed)
+        rng = np.random.default_rng(seed=self.config.random_seed)
 
         data = self.get_data('input')
         with open(self.config.settings_file, 'rb') as handle:
@@ -189,13 +189,13 @@ def run(self):
                 number_to_keep = len(temp_data)
 
             if int(number_to_keep) != number_to_keep:
-                random_num = np.random.uniform()
+                random_num = rng.uniform()
             else:
                 random_num = 2
 
             number_to_keep = np.floor(number_to_keep)
             indices_to_list = list(temp_data.index.values)
-            np.random.shuffle(indices_to_list)
+            rng.shuffle(indices_to_list)
 
             if random_num > xratio:  # pragma: no cover
                 for j in range(int(number_to_keep)):

src/rail/estimation/algos/knnpz.py

@@ -96,8 +96,8 @@ def run(self):
         trainszs = np.array(training_data[self.config.redshift_col])
         colordata = _computecolordata(knndf, self.config.ref_band, self.config.bands)
         nobs = colordata.shape[0]
-        rng = np.random.default_rng
-        perm = rng().permutation(nobs)
+        rng = np.random.default_rng(seed=self.config.seed)
+        perm = rng.permutation(nobs)
         ntrain = round(nobs * self.config.trainfrac)
         xtrain_data = colordata[perm[:ntrain]]
         train_data = copy.deepcopy(xtrain_data)

src/rail/estimation/algos/randomPZ.py

@@ -23,6 +23,7 @@ class RandomPZ(CatEstimator):
                           rand_zmin=Param(float, 0.0, msg="The minimum redshift of the z grid"),
                           rand_zmax=Param(float, 3.0, msg="The maximum redshift of the z grid"),
                           nzbins=Param(int, 301, msg="The number of gridpoints in the z grid"),
+                          seed=Param(int, 87, msg="random seed"),
                           column_name=Param(str, "mag_i_lsst", msg="name of a column that has the correct number of galaxies to find length of"))
 
     def __init__(self, args, comm=None):
@@ -35,7 +36,8 @@ def _process_chunk(self, start, end, data, first):
         pdf = []
         # allow for either format for now
         numzs = len(data[self.config.column_name])
-        zmode = np.round(np.random.uniform(0.0, self.config.rand_zmax, numzs), 3)
+        rng = np.random.default_rng(seed=self.config.seed + start)
+        zmode = np.round(rng.uniform(0.0, self.config.rand_zmax, numzs), 3)
         widths = self.config.rand_width * (1.0 + zmode)
         self.zgrid = np.linspace(self.config.rand_zmin, self.config.rand_zmax, self.config.nzbins)
         for i in range(numzs):