1D triple well dataset is broken

[pietronvll]

The 1D Triple Well potential

kooplearn/kooplearn/datasets/stochastic.py

Line 338 in ca71864

class LangevinTripleWell1D(DiscreteTimeDynamics):

is bugged as:

• The .npz resource files are not being included in the pip installer, and should be loaded using importlib.resources
• The Boltzmann PDF

  kooplearn/kooplearn/datasets/stochastic.py

  Line 399 in ca71864

  boltzmann_pdf = (dx**-1) * scipy.special.softmax(self.force_fn(cc) / self.kt)

  is wrong, and it should be

  boltzmann_pdf = (dx**-1) * scipy.special.softmax(-self.potential_fn(cc) / self.kt)

  with

  def potential_fn(self, x: ArrayLike) -> ArrayLike:
     """
     See Example 1 of "Modeling Molecular Kinetics with tICA and the Kernel Trick" 10.1021/ct5007357
     """
     return 4 * (
         x**8
         + 0.8 * np.exp(-80 * (x**2))
         + 0.2 * np.exp(-80 * ((x - 0.5) ** 2))
         + 0.5 * np.exp(-40 * ((x + 0.5) ** 2))
     )