check output for finite data and raise a useful error message

Fixes #1272

pyemma/datasets/api.py

@@ -135,7 +135,7 @@ def get_quadwell_data(ntraj=10, nstep=10000, x0=0., nskip=1, dt=0.01, kT=1.0, ma
         number of integrator steps
     dt: float, default=0.01
         time step size
-    kT: float, default=10.0
+    kT: float, default=1.0
         temperature factor
     mass: float, default=1.0
         mass
@@ -149,5 +149,11 @@ def get_quadwell_data(ntraj=10, nstep=10000, x0=0., nskip=1, dt=0.01, kT=1.0, ma
     """
     from .potentials import PrinzModel
     pw = PrinzModel(dt, kT, mass=mass, damping=damping)
-    trajs = [pw.sample(x0, nstep, nskip=nskip) for _ in range(ntraj)]
+    import warnings
+    import numpy as np
+    with warnings.catch_warnings(record=True) as w:
+        trajs = [pw.sample(x0, nstep, nskip=nskip) for _ in range(ntraj)]
+        if not np.all(tuple(np.isfinite(x) for x in trajs)):
+            raise RuntimeError('integrator detected invalid values in output. If you used a high temperature value (kT),'
+                               ' try decreasing the integration time step dt.')
     return trajs

pyemma/datasets/tests/test_datasets.py

@@ -20,6 +20,7 @@
 from pyemma.thermo import estimate_umbrella_sampling
 from pyemma.thermo import estimate_multi_temperature
 from numpy.testing import assert_allclose
+import unittest
 
 
 def test_umbrella_sampling_data():
@@ -82,6 +83,14 @@ def test_multi_temperature_data():
     assert_allclose(pi, [0.3, 0.7], rtol=0.25, atol=0.1)
 
 
-def test_prinz_potential():
-    from pyemma.datasets import get_quadwell_data
-    get_quadwell_data()
+class TestPrinzPotential(unittest.TestCase):
+
+    def test_prinz_potential(self):
+        from pyemma.datasets import get_quadwell_data
+        import numpy as np
+
+        d = get_quadwell_data(ntraj=1, nstep=int(1e5))
+        assert np.all(np.isfinite(x) for x in d)
+
+        with self.assertRaises(RuntimeError):
+            get_quadwell_data(ntraj=1, dt=1, kT=100)