.. _MITViterbi:

MITViterbi
==========

.. plugin:: MITViterbi(FramePass)

    Plugin Type: :py:class:`~diplomat.predictors.fpe.frame_pass.FramePass`

    An implementation of the Multi-Individual Tracking Viterbi algorithm. Runs a viterbi-like algorithm across the frames to determine the maximum scoring paths per individual, assuming an individuals can't take a paths that would have been more likely for other individuals to have taken.

    .. rubric:: Settings


    .. py:setting:: standard_deviation
        :type: Union[float, Literal['auto']]
        :value: 'auto'

    The standard deviation of the 2D Gaussian curve used for transition probabilities. Defaults to 'auto', which attempts to use an optimized value if one has been computed by a prior pass, and otherwise uses the default value of 1...

    .. py:setting:: skeleton_weight
        :type: float
        :value: 1

    A positive float, determines how much impact probabilities from skeletal transitions should have in each forward/backward step if a skeleton was created and enabled by prior passes... This is not a probability, but rather a ratio.

    .. py:setting:: minimum_skeleton_weight
        :type: float
        :value: 0.0001

    A positive float, bounds skeleton_weight from below in the adjustment of skeleton weight for poor fix frame quality. Avoiding a zero skeleton weight is necessary to stop information loss across segment boundaries in SFPE.

    .. py:setting:: soft_domination_weight
        :type: float
        :value: 1

    A positive float, determines how much impact probabilities from soft domination transitions should have in each forward/backward step if soft domination was enabled This is not a probability, but rather a ratio.

    .. py:setting:: soft_domination_spread
        :type: float
        :value: 3

    A positive float, the standard deviation of the viterbi is multiplied by this value to determine the standard deviation of the soft domination gaussian.

    .. py:setting:: amplitude
        :type: float
        :value: 1

    The max amplitude of the 2D Gaussian curve used for transition probabilities.

    .. py:setting:: lowest_value
        :type: float
        :value: 0

    The lowest value the 2D Gaussian curve used for transition probabilities can reach.

    .. py:setting:: obscured_probability
        :type: RangedFloat[min=0.0, max=1.0]
        :value: 1e-06

    A constant float between 0 and 1 that determines the prior probability of being in any hidden state cell.

    .. py:setting:: minimum_obscured_probability
        :type: RangedFloat[min=0.0, max=1.0]
        :value: 1e-12

    A constant float between 0 and 1 that sets a cutoff for obscured state probabilities.

    .. py:setting:: enter_state_probability
        :type: RangedFloat[min=0.0, max=1.0]
        :value: 1e-12

    A constant, the probability of being in the enter state.

    .. py:setting:: enter_state_exit_probability
        :type: RangedFloat[min=0.0, max=1.0]
        :value: 0.9999

    A constant, the probability of exiting the enter state. Probability of staying in the enter state is this value subtracted from 1.

    .. py:setting:: obscured_survival_max
        :type: int
        :value: 50

    An integer, the max number of points to allow to survive for each frame, if there is more than this value, the top ones are kept.

    .. py:setting:: obscured_decay_rate
        :type: RangedFloat[min=0.0, max=1.0]
        :value: 0.99

    A constant float defining the decay rate of probabilities in the occluded state.

    .. py:setting:: gaussian_plateau
        :type: Union[float, Literal[None]]
        :value: None

    A float specifying the area over which to flatten the gaussian curve should be less than the norm_dist value. If none, set to the norm_dist.

    .. py:setting:: include_skeleton
        :type: bool
        :value: True

    A boolean. If True, include skeleton information in the forward backward pass, otherwise don't. If no skeleton has been built in a prior pass, does nothing.

    .. py:setting:: include_soft_domination
        :type: bool
        :value: False

    A boolean, if True, enable soft domination in MIT-Viterbi algorithm.Otherwise soft domination probabilities are excluded.

    .. py:setting:: square_distances
        :type: bool
        :value: False

    A boolean. If True, square distances between points before putting them through the gaussian, otherwise don't.

    .. py:setting:: lowest_skeleton_score
        :type: RangedFloat[min=-inf, max=0.0]
        :value: -inf

    A float, the lowest allowed log-probability for the distribution of skeleton scores.This prevents the skeleton transitions from zeroing all probabilities in the viterbi.