Skip to content

comparison

Set of helpful functions for comparing different state discretizations.

BSD 3-Clause License Copyright © 2019-2023, Daniel Nagel All rights reserved.

compare_discretization(traj1, traj2, method='symmetric')

Compare similarity of two state discretizations.

This method compares the similarity of two state discretizations of the same dataset. There are two different methods, 'directed' gives a measure on how high is the probable to assign a frame correclty knowing the traj1. Hence splitting a state into many is not penalized, while merging multiple into a single state is. Selecting 'symmetric' it is check in both directions, so it checks for each state if it is possible to assigned it forward or backward. Hence, splitting and merging states is not penalized.

Parameters:

  • traj1 (StateTraj like) –

    First state discretization.

  • traj2 (StateTraj like) –

    Second state discretization.

  • method ([symmetric, directed], default: 'symmetric' ) –

    Selecting similarity norm. 'symmetric' compares if each frame is forward or backward assignable, while 'directed' checks only if it is forard assignable.

Returns:

  • similarity ( float ) –

    Similarity going from [0, 1], where 1 means identical and 0 no similarity at all.

Source code in src/msmhelper/md/comparison.py 
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
def compare_discretization(traj1, traj2, method='symmetric'):
    """Compare similarity of two state discretizations.

    This method compares the similarity of two state discretizations of the
    same dataset. There are two different methods, 'directed' gives a measure
    on how high is the probable to assign a frame correclty knowing the
    `traj1`. Hence splitting a state into many is not penalized, while merging
    multiple into a single state is. Selecting 'symmetric' it is check in both
    directions, so it checks for each state if it is possible to assigned it
    forward or backward. Hence, splitting and merging states is not penalized.

    Parameters
    ----------
    traj1 : StateTraj like
        First state discretization.
    traj2 : StateTraj like
        Second state discretization.
    method : ['symmetric', 'directed']
        Selecting similarity norm. 'symmetric' compares if each frame is
        forward or backward assignable, while 'directed' checks only if it is
        forard assignable.

    Returns
    -------
    similarity : float
        Similarity going from [0, 1], where 1 means identical and 0 no
        similarity at all.

    """
    # format input
    traj1, traj2 = StateTraj(traj1), StateTraj(traj2)
    if method not in {'symmetric', 'directed'}:
        raise ValueError(
            'Only methods "symmetric" and "directed" are supported',
        )

    # check if same length
    if traj1.nframes != traj2.nframes:
        raise ValueError(
            'Trajectories are of different length: ' +
            '{0} vs {1}'.format(traj1.nframes, traj2.nframes),
        )

    # check if only single state
    if traj1.nstates == 1 or traj2.nstates == 1:
        raise ValueError(
            'Trajectories needs to have at least two states: ' +
            '{0} and {1}'.format(traj1.nstates, traj2.nstates),
        )
    return _compare_discretization(traj1, traj2, method)