from scipy.stats import multinomial
from ..base import Property
from ..models.measurement.categorical import MarkovianMeasurementModel
from ..sensor.sensor import Sensor
from ..types.array import StateVector
from ..types.detection import TrueCategoricalDetection
[docs]
class HMMSensor(Sensor):
r"""Sensor model that observes a categorical state space and returns categorical measurements.
Measurements are categorical distributions over a finite set of categories
:math:`Z = \{\zeta^n|n\in \mathbf{N}, n\le N\}` (for some finite :math:`N`).
"""
measurement_model: MarkovianMeasurementModel = Property(
doc="Measurement model to generate detection vectors from"
)
@property
def ndim_state(self):
return self.measurement_model.ndim_state
@property
def ndim_meas(self):
return self.measurement_model.ndim_meas
[docs]
def measure(self, ground_truths, noise: bool = True, **kwargs):
r"""Generate a categorical measurement for a given set of true categorical state.
Parameters
----------
ground_truths: Set[:class:`~.CategoricalGroundTruthState`]
A set of :class:`~.CategoricalGroundTruthState`.
noise: bool
Indicates whether measurement vectors are sampled from and the resultant measurement
categories returned instead. These are discrete categories instead of a distribution
over the measurement space. They are represented by N-tuples, with all components
equal to 0, except at an index corresponding to the relevant category.
For example :math:`e^k` indicates that the measurement category is :math:`\zeta^k`.
If `False`, the resultant distribution is returned.
Returns
-------
Set[:class:`~.TrueCategoricalDetection`]
A set of measurements generated from the given states. The timestamps of the
measurements are set equal to that of the corresponding states that they were
calculated from. Each measurement stores the ground truth path that it was produced
from.
"""
detections = set()
for truth in ground_truths:
timestamp = truth.timestamp
detection_vector = self.measurement_model.function(truth, noise=noise, **kwargs)
if noise:
# Sample from resultant distribution
rv = multinomial(n=1, p=detection_vector.flatten())
detection_vector = StateVector(rv.rvs(size=1, random_state=None))
detection = TrueCategoricalDetection(
state_vector=detection_vector,
timestamp=timestamp,
categories=self.measurement_model.measurement_categories,
measurement_model=self.measurement_model,
groundtruth_path=truth
)
detections.add(detection)
return detections