models#

perceptivo.psychophys.model.f_to_bark(frequency: float) → float#

Convert frequency to Bark using [WSG91]

Parameters: frequency (float) – Frequency to convert
Returns: (float) Bark

perceptivo.psychophys.model.bark_to_f(bark: float) → float#

Convert bark to frequency using inverted [WSG91]

Parameters: bark (float) – bark to convert
Returns: (float) frequency

class perceptivo.psychophys.model.Audiogram_Model(freq_range: Tuple[float, float] = (125, 8500), amplitude_range: Tuple[float, float] = (5, 60), exam_params: Optional[perceptivo.types.exam.Exam_Params] = None, *args, **kwargs)#

Bases: perceptivo.root.Perceptivo_Object

Metaclass for Audiogram models and estimators.

These classes are used to estimate the audiogram, as well as control the order of the presentation of probe sounds.

Note

This class may be split into an experimental runner class and an audiogram model, but since the choice of the next stimulus should ideally be based on the current audiogram model, they are built together for now.

Parameters

freq_range (tuple) – Tuple of two floats indicating min/max frequency (default: (125, 8500))
amplitude_range (tuple) – Tuple of two floats indicating min/max amplitude in dbSPL (default: (5,60))

Variables

audiogram (types.psychophys.Audiogram) – Audiogram of model
samples (types.psychophys.Samples) – Individual samples of frequency/amplitude and whether a sound was detected.

abstract update(sample: perceptivo.types.psychophys.Sample)#: Update the model with a new :class:`~.types.psychophys.Sample

abstract next() → perceptivo.types.sound.Sound#

Generate parameters for the next Sound to be presented

Next should generate samples that respect the frequencies and amplitudes set in exam_params, if present. As well as allow_repeats

class perceptivo.psychophys.model.Gaussian_Process(kernel: Optional[Union[sklearn.gaussian_process.kernels.Kernel, perceptivo.types.psychophys.Kernel]] = None, *args, **kwargs)#

Bases: perceptivo.psychophys.model.Audiogram_Model

Gaussian process model based on [CdeVries16]

Model: * Bayesian Process Classifier, predicting binary audibility as a function of frequency and amplitude * Kernel: * Covariance Function: Squared Exponent (RBF)

Process: * Convert sampled frequency to bark with f_to_bark() * Update model * Generate next stimulus * Convert back to freq

Examples

from perceptivo.psychophys.oracle import reference_audiogram
from perceptivo.psychophys.model import Gaussian_Process
from perceptivo.types.psychophys import Sample

oracle = reference_audiogram(scale=3)
model = Gaussian_Process(amplitude_range=(5,35))

for i in range(100):
    sound = model.next()
    sample = Sample(response=oracle(sound), sound=sound)
    model.update(sample)

model.plot()

(Source code, png, hires.png, pdf)

References

[CdeVries16]
[GMG+15]
[MSG16]
[]

property kernel: sklearn.gaussian_process.kernels.Kernel#

Kernel used in the gaussian process model. If None is given on init, use the types.psychophys.Kernel

Returns: sklearn.gaussian_process.kernels.Kernel

property samples: perceptivo.types.psychophys.Samples#

Stored samples from updates

Returns: Samples

update(sample: perceptivo.types.psychophys.Sample)#

Update the model with a new sample!

Parameters: sample ()

_get_params() → Tuple[float, float]#

Generate sound params

Returns: a tuple of freq, amp

next() → perceptivo.types.sound.Sound#

Generate parameters for the next sound to present

Returns: Sound

plot(mesh_resolution: int = 5)#

References

https://scikit-learn.org/stable/auto_examples/gaussian_process/plot_gpc_iris.html#sphx-glr-auto-examples-gaussian-process-plot-gpc-iris-py

Returns: