AnalysisBase.run() misinterpretation of NumPy booleans frame masks as integer frame indices

[sici17]

Expected behavior

AnalysisBase.run() should interpret Python and NumPy boolean masks in the same way. For example:

frames = [True, False, True, False] AND frames = np.array([True, False, True, False], dtype=bool)

should equally select as output frames [0, 2].

Actual behavior

The case with the Python boolean list correctly selects frames [0, 2] as we expect, but an equivalent NumPy boolean array misinterpret them as integer frame indices and processes [1, 0, 1, 0].

This happens without throwing any error and can produce plausible-looking analysis results calculated from the wrong frames.

Code to reproduce the behavior

import numpy as np
import MDAnalysis as mda

from MDAnalysis.analysis.base import AnalysisFromFunction


# We create a trajectory with one atom and four frames.
coordinates = np.zeros((4, 1, 3), dtype=np.float32)

# Encode the frame number in the atom's x coordinate:
coordinates[:, 0, 0] = [0, 1, 2, 3]

u = mda.Universe.empty(n_atoms=1)
u.load_new(coordinates, format="MEMORY")


def read_x(atomgroup):
    """Return the x coordinate of the only atom."""
    return atomgroup.positions[0, 0]


# Both masks should select frames 0 and 2.
python_mask = [True, False, True, False]
numpy_mask = np.array([True, False, True, False], dtype=bool)

with_python_mask = AnalysisFromFunction(
    read_x,
    u.trajectory,
    u.atoms,
).run(frames=python_mask)

with_numpy_mask = AnalysisFromFunction(
    read_x,
    u.trajectory,
    u.atoms,
).run(frames=numpy_mask)


print("Python list values:", with_python_mask.results.timeseries)
print("NumPy array values:", with_numpy_mask.results.timeseries)

Output:

Python list values: [0. 2.]
NumPy array values: [1. 0. 1. 0.]

Both inputs are boolean masks and should select frames [0, 2]. However, the NumPy mask is interpreted as the integer frame indices [1, 0, 1, 0].

Current version of MDAnalysis

-MDAnalysis: 2.10.0
This behavior was originally noted by @srikarjy while working on #5459.