Use Numpy rng in FDEM tests (#1449)

Replace the usage of the deprecated functions in `numpy.random` module
for the Numpy's random number generator class and its methods, in most
of the FDEM tests. Increase number of iterations for checking
derivatives where needed.

Part of the solution to #1289

---------

Co-authored-by: Lindsey Heagy <lindseyheagy@gmail.com>

tests/em/fdem/forward/test_FDEM_casing.py

@@ -12,9 +12,10 @@
 sigma = np.r_[10.0, 5.5e6, 1e-1]
 mu = mu_0 * np.r_[1.0, 100.0, 1.0]
 srcloc = np.r_[0.0, 0.0, 0.0]
-xobs = np.random.rand(n) + 10.0
+rng = np.random.default_rng(seed=42)
+xobs = rng.uniform(size=n) + 10.0
 yobs = np.zeros(n)
-zobs = np.random.randn(n)
+zobs = rng.normal(size=n)
 
 
 def CasingMagDipoleDeriv_r(x):
@@ -63,15 +64,24 @@ def CasingMagDipole2Deriv_z_z(z):
 
 class Casing_DerivTest(unittest.TestCase):
     def test_derivs(self):
+        rng = np.random.default_rng(seed=42)
+
+        np.random.seed(1983)  # set a random seed for check_derivative
         tests.check_derivative(
-            CasingMagDipoleDeriv_r, np.ones(n) * 10 + np.random.randn(n), plotIt=False
+            CasingMagDipoleDeriv_r, np.ones(n) * 10 + rng.normal(size=n), plotIt=False
         )
-        tests.check_derivative(CasingMagDipoleDeriv_z, np.random.randn(n), plotIt=False)
+
+        np.random.seed(1983)  # set a random seed for check_derivative
+        tests.check_derivative(CasingMagDipoleDeriv_z, rng.normal(size=n), plotIt=False)
+
+        np.random.seed(1983)  # set a random seed for check_derivative
         tests.check_derivative(
             CasingMagDipole2Deriv_z_r,
-            np.ones(n) * 10 + np.random.randn(n),
+            np.ones(n) * 10 + rng.normal(size=n),
             plotIt=False,
         )
+
+        np.random.seed(1983)  # set a random seed for check_derivative
         tests.check_derivative(
-            CasingMagDipole2Deriv_z_z, np.random.randn(n), plotIt=False
+            CasingMagDipole2Deriv_z_z, rng.normal(size=n), plotIt=False
         )

tests/em/fdem/forward/test_FDEM_primsec.py

@@ -16,8 +16,6 @@
 TOL_JT = 1e-10
 FLR = 1e-20  # "zero", so if residual below this --> pass regardless of order
 
-np.random.seed(2016)
-
 # To test the primary secondary-source, we look at make sure doing primary
 # secondary for a simple model gives comprable results to just solving a 3D
 # problem
@@ -128,15 +126,17 @@ def fun(x):
                 lambda x: self.secondarySimulation.Jvec(x0, x, f=self.fields_primsec),
             ]
 
+        np.random.seed(1983)  # set a random seed for check_derivative
         return tests.check_derivative(fun, x0, num=2, plotIt=False)
 
     def AdjointTest(self):
         print("\nTesting adjoint")
 
         m = model
         f = self.fields_primsec
-        v = np.random.rand(self.secondarySurvey.nD)
-        w = np.random.rand(self.secondarySimulation.sigmaMap.nP)
+        rng = np.random.default_rng(seed=2016)
+        v = rng.uniform(size=self.secondarySurvey.nD)
+        w = rng.uniform(size=self.secondarySimulation.sigmaMap.nP)
 
         vJw = v.dot(self.secondarySimulation.Jvec(m, w, f))
         wJtv = w.dot(self.secondarySimulation.Jtvec(m, v, f))

tests/em/fdem/forward/test_properties.py

@@ -44,12 +44,13 @@ def test_source_properties_validation():
     # LineCurrent
     with pytest.raises(TypeError):
         fdem.sources.LineCurrent([], frequency, location=["a", "b", "c"])
+    rng = np.random.default_rng(seed=42)
+    random_locations = rng.normal(size=(5, 3, 2))
     with pytest.raises(ValueError):
-        fdem.sources.LineCurrent([], frequency, location=np.random.rand(5, 3, 2))
+        fdem.sources.LineCurrent([], frequency, location=random_locations)
+    random_locations = rng.normal(size=(5, 3))
     with pytest.raises(ValueError):
-        fdem.sources.LineCurrent(
-            [], frequency, location=np.random.rand(5, 3), current=0.0
-        )
+        fdem.sources.LineCurrent([], frequency, location=random_locations, current=0.0)
 
 
 def test_bad_source_type():

tests/em/fdem/inverse/adjoint/test_FDEM_adjointEB.py

@@ -26,17 +26,18 @@ def adjointTest(fdemType, comp):
     m = np.log(np.ones(prb.sigmaMap.nP) * CONDUCTIVITY)
     mu = np.ones(prb.mesh.nC) * MU
 
+    rng = np.random.default_rng(seed=42)
     if addrandoms is True:
-        m = m + np.random.randn(prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
-        mu = mu + np.random.randn(prb.mesh.nC) * MU * 1e-1
+        m = m + rng.normal(size=prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
+        mu = mu + rng.normal(size=prb.mesh.nC) * MU * 1e-1
 
     survey = prb.survey
     # prb.PropMap.PropModel.mu = mu
     # prb.PropMap.PropModel.mui = 1./mu
     u = prb.fields(m)
 
-    v = np.random.rand(survey.nD)
-    w = np.random.rand(prb.mesh.nC)
+    v = rng.uniform(size=survey.nD)
+    w = rng.uniform(size=prb.mesh.nC)
 
     vJw = v.dot(prb.Jvec(m, w, u))
     wJtv = w.dot(prb.Jtvec(m, v, u))

tests/em/fdem/inverse/adjoint/test_FDEM_adjointHJ.py

@@ -26,15 +26,16 @@ def adjointTest(fdemType, comp, src):
     m = np.log(np.ones(prb.sigmaMap.nP) * CONDUCTIVITY)
     mu = np.ones(prb.mesh.nC) * MU
 
+    rng = np.random.default_rng(seed=42)
     if addrandoms is True:
-        m = m + np.random.randn(prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
-        mu = mu + np.random.randn(prb.mesh.nC) * MU * 1e-1
+        m = m + rng.normal(size=prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
+        mu = mu + rng.normal(size=prb.mesh.nC) * MU * 1e-1
 
     survey = prb.survey
     u = prb.fields(m)
 
-    v = np.random.rand(survey.nD)
-    w = np.random.rand(prb.mesh.nC)
+    v = rng.uniform(size=survey.nD)
+    w = rng.uniform(size=prb.mesh.nC)
 
     vJw = v.dot(prb.Jvec(m, w, u))
     wJtv = w.dot(prb.Jtvec(m, v, u))

tests/em/fdem/inverse/derivs/test_FDEM_derivs.py

@@ -29,19 +29,18 @@
 
 def derivTest(fdemType, comp, src):
     prb = getFDEMProblem(fdemType, comp, SrcType, freq)
-    # prb.solverOpts = dict(check_accuracy=True)
 
     print(f"{fdemType} formulation {src} - {comp}")
     x0 = np.log(np.ones(prb.sigmaMap.nP) * CONDUCTIVITY)
-    # mu = np.log(np.ones(prb.mesh.nC)*MU)
 
     if addrandoms is True:
-        x0 = x0 + np.random.randn(prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
-        # mu = mu + np.random.randn(prb.sigmaMap.nP)*MU*1e-1
+        rng = np.random.default_rng(seed=42)
+        x0 = x0 + rng.normal(size=prb.sigmaMap.nP) * np.log(CONDUCTIVITY) * 1e-1
 
     def fun(x):
         return prb.dpred(x), lambda x: prb.Jvec(x0, x)
 
+    np.random.seed(1983)  # set a random seed for check_derivative
     return tests.check_derivative(fun, x0, num=2, plotIt=False, eps=FLR)
 
 

tests/em/fdem/muinverse/test_muinverse.py

@@ -18,8 +18,9 @@ def setupMeshModel():
     hz = [(cs, npad, -1.3), (cs, nc), (cs, npad, 1.3)]
 
     mesh = discretize.CylindricalMesh([hx, 1.0, hz], "0CC")
-    muMod = 1 + MuMax * np.random.randn(mesh.nC)
-    sigmaMod = np.random.randn(mesh.nC)
+    rng = np.random.default_rng(seed=2016)
+    muMod = 1 + MuMax * rng.normal(size=mesh.nC)
+    sigmaMod = rng.normal(size=mesh.nC)
 
     return mesh, muMod, sigmaMod
 
@@ -149,7 +150,8 @@ def test_mats_cleared(self):
         MfMuiIDeriv_zero = self.simulation.MfMuiIDeriv(utils.Zero())
         MeMuDeriv_zero = self.simulation.MeMuDeriv(utils.Zero())
 
-        m1 = np.random.rand(self.mesh.nC)
+        rng = np.random.default_rng(seed=2016)
+        m1 = rng.uniform(size=self.mesh.nC)
         self.simulation.model = m1
 
         self.assertTrue(getattr(self, "_MeMu", None) is None)
@@ -168,7 +170,6 @@ def JvecTest(
         self.setUpProb(prbtype, sigmaInInversion, invertMui)
         print("Testing Jvec {}".format(prbtype))
 
-        np.random.seed(3321)
         mod = self.m0
 
         def fun(x):
@@ -177,19 +178,21 @@ def fun(x):
                 lambda x: self.simulation.Jvec(mod, x),
             )
 
-        dx = np.random.rand(*mod.shape) * (mod.max() - mod.min()) * 0.01
+        rng = np.random.default_rng(seed=3321)
+        dx = rng.uniform(size=mod.shape) * (mod.max() - mod.min()) * 0.01
 
-        return tests.check_derivative(fun, mod, dx=dx, num=3, plotIt=False)
+        np.random.seed(1983)  # set a random seed for check_derivative
+        return tests.check_derivative(fun, mod, dx=dx, num=4, plotIt=False)
 
     def JtvecTest(
         self, prbtype="ElectricField", sigmaInInversion=False, invertMui=False
     ):
         self.setUpProb(prbtype, sigmaInInversion, invertMui)
         print("Testing Jvec {}".format(prbtype))
 
-        np.random.seed(31345)
-        u = np.random.rand(self.simulation.muMap.nP)
-        v = np.random.rand(self.survey.nD)
+        rng = np.random.default_rng(seed=3321)
+        u = rng.uniform(size=self.simulation.muMap.nP)
+        v = rng.uniform(size=self.survey.nD)
 
         self.simulation.model = self.m0