pulse2percept.utils¶

Various utility and helper functions.

`base`	`PrettyPrint`, `Frozen`, `Data`, `bijective26_name`, `cached`, `gamma`, `unique`
`constants`	`DT`, `MIN_AMP`, `VIDEO_BLOCK_SIZE`, `ZORDER`
`geometry`	`Grid2D`, `VisualFieldMap`, `Curcio1990Map`, `Watson2014Map`, `Watson2014DisplaceMap`, `cart2pol`, `pol2cart`, `delta_angle`
`array`	`is_strictly_increasing`, `sample`, `unique`, `radial_mask`
`images`	`center_image`, `scale_image`, `shift_image`, `trim_image`
`convolution`	`conv`, `center_vector`
`optimize`	`bisect`
`stats`	`r2_score`, `circ_r2_score`
`parallel`	`parfor`
`deprecation`	`deprecated`, `is_deprecated`

pulse2percept.utils.bijective26_name(i)[source]¶

Bijective base-26 numeration

Creates the “alphabetic number” for a given integer i following bijective base-26 numeration: A-Z, AA-AZ, BA-BZ, … ZA-ZZ, AAA-AAZ, ABA-ABZ, …

Parameters:	i (int) – Regular number to be translated into an alphabetic number
Returns:	name (string) – Alphabetic number

Examples

>>> bijective26_name(0)
'A'

>>> bijective26_name(26)
'AA'

pulse2percept.utils.bisect(y_target, func, args=None, kwargs=None, x_lo=0, x_hi=1, x_tol=1e-06, y_tol=0.001, max_iter=100)[source]¶

Binary search (bisection method) to find x value that gives y_target

For a function y = func(x, *args, **kwargs), returns x_opt for which func(x_opt, *args, **kwargs) is approximately equal to y_target.

New in version 0.7.

Parameters:

y_target (float) – Target y value
args, kwargs (func,) – The function to call along with its positional and keyword arguments
x_hi (x_lo,) – Lower and upper bounds on x
x_tol (float, optional) – Search will stop if the range of candidate x values is smaller than x_tol
y_tol (float, optional) – Search will stop if y is within y_tol of y_target
max_iter (int, optional) – Maximum number of iterations to run

Returns:

x_opt (float) – The x value such that func(x_opt) $approx$ y_target

Notes

Assumes func is a monotonously increasing function of x.
Does not require x_lo and x_hi to have opposite signs as in the conventional bisection method.

pulse2percept.utils.cached(f)[source]¶

Cached property decorator

Decorator can be added to the property of a class to maintain a cache. This is useful when computing the property is computationall expensive. The property will only be computed on first call, and subsequent calls will refer to the cached result.

Important

When making use of a cached property, the class should also maintain a _cache_active flag set to True or False.

New in version 0.7.

pulse2percept.utils.cart2pol(x, y)[source]¶

Convert Cartesian to polar coordinates

Parameters:	y (x,) – The x,y Cartesian coordinates
Returns:	theta, rho (scalar or array-like) – The transformed polar coordinates

pulse2percept.utils.center_image(img, loc=None)[source]¶

Center the image foreground

This function shifts the center of mass (CoM) to the image center. The background of the image is assumed to be black (0 grayscale).

New in version 0.7.

Parameters:	img (ndarray) – A 2D NumPy array representing a (height, width) grayscale image, or a 3D NumPy array representing a (height, width, channels) RGB image loc ((col, row), optional) – The pixel location at which to center the CoM. By default, shifts the CoM to the image center.
Returns:	img (ndarray) – A copy of the image centered at `loc`

pulse2percept.utils.center_vector(vec, newlen)[source]¶

Returns the center newlen portion of a vector.

Adapted from scipy.signal.signaltools._centered: github.com/scipy/scipy/blob/v0.18.0/scipy/signal/signaltools.py#L236-L243

pulse2percept.utils.circ_r2_score(y_true, y_pred)[source]¶

Calculate circular R² (the coefficient of determination)

The best possible score is 1.0, lower values are worse.

New in version 0.7.

Parameters:	y_true (array-like) – Ground truth (correct) target values. y_pred (array-like) – Estimated target values.
Returns:	z (float) – The R² score

Notes

If the ground-truth data has zero variance, R² will be zero.
This is not a symmetric function

pulse2percept.utils.conv(data, kernel, mode='full', method='fft')[source]¶

Convoles data with a kernel using either FFT or sparse convolution

This function convolves data with a kernel, relying either on the fast Fourier transform (FFT) or a sparse convolution function.

Parameters:

data (array_like) – First input, typically the data array
kernel (array_like) – Second input, typically the kernel
mode (str {'full', 'valid', 'same'}, optional, default: 'full') –
A string indicating the size of the output:
- full:
  
  The output is the full discrete linear convolution of the inputs.
- valid:
  
  The output consists only of those elements that do not rely on zero-padding.
- same:
  
  The output is the same size as data, centered with respect to the ‘full’ output.
method (str {'fft', 'sparse'}, optional, default: 'fft') –
A string indicating the convolution method:
- fft:
  
  Use the fast Fourier transform (FFT).
- sparse:
  
  Use the sparse convolution.

class pulse2percept.utils.Curcio1990Map[source]¶

Converts between visual angle and retinal eccentricity [Curcio1990]

static dva2ret(xdva, ydva)[source]¶

Convert degrees of visual angle (dva) to retinal eccentricity (um)

Assumes that one degree of visual angle is equal to 280 um on the retina [Curcio1990].

static ret2dva(xret, yret)[source]¶

Convert retinal eccentricity (um) to degrees of visual angle (dva)

Assumes that one degree of visual angle is equal to 280 um on the retina [Curcio1990]

class pulse2percept.utils.Data(data, axes=None, metadata=None)[source]¶

N-dimensional data container

New in version 0.6.

Parameters:	data (np.ndarray) – An N-dimensional NumPy array containing the data to store axes (dict or tuple, optional) – For each dimension in `data`, specify axis name and labels. metadata (dict, optional) – A dictionary that can store arbitrary metadata

pulse2percept.utils.delta_angle(source_angle, target_angle, hi=6.283185307179586)[source]¶

Returns the signed difference between two angles (rad)

The difference is calculated as target_angle - source_angle. The difference will thus be positive if target_angle > source_angle.

New in version 0.7.

Parameters:	target_angle (source_angle,) – Input arrays with circular data in the range [0, hi] hi (float, optional) – Sets the upper bounds of the range (e.g., 2*np.pi or 360). Lower bound is always 0
Returns:	The signed difference target_angle - source_angle in [0, hi]

class pulse2percept.utils.deprecated(alt_func=None, deprecated_version=None, removed_version=None)[source]¶

Decorator to mark deprecated functions and classes with a warning.

Parameters:	alt_func (str) – If given, tell user what function to use instead. deprecated_version (float or str) – The package version in which the function/class was first marked as deprecated. removed_version (float or str) – The package version in which the deprecated function/class will be removed.

exception pulse2percept.utils.FreezeError[source]¶

Exception class used to raise when trying to add attributes to Frozen Classes of type Frozen do not allow for new attributes to be set outside the constructor.

with_traceback()¶: Exception.with_traceback(tb) – set self.__traceback__ to tb and return self.

class pulse2percept.utils.Frozen[source]¶: “Frozen” classes (and subclasses) do not allow for new class attributes to be set outside the constructor. On attempting to add a new attribute, the class will raise a FreezeError.

pulse2percept.utils.gamma(n, tau, tsample, tol=0.01)[source]¶

Returns the impulse response of n cascaded leaky integrators

This function calculates the impulse response of n cascaded leaky integrators with constant of proportionality 1/tau: y = (t/theta).^(n-1).*exp(-t/theta)/(theta*factorial(n-1))

Parameters:	n (int) – Number of cascaded leaky integrators tau (float) – Decay constant of leaky integration (seconds). Equivalent to the inverse of the constant of proportionality. tsample (float) – Sampling time step (seconds). tol (float) – Cut the kernel to size by ignoring function values smaller than a fraction `tol` of the peak value.

class pulse2percept.utils.Grid2D(x_range, y_range, step=1, grid_type='rectangular')[source]¶

2D spatial grid

This class generates a two-dimensional mesh grid from a range of x, y values and provides an iterator to loop over elements.

New in version 0.6.

Parameters:

x_range ((x_min, x_max)) – A tuple indicating the range of x values (includes end points)
y_range (tuple, (y_min, y_max)) – A tuple indicating the range of y values (includes end points)
step (int, double, tuple) – Step size. If int or double, the same step will apply to both x and y ranges. If a tuple, it is interpreted as (x_step, y_step).
grid_type ({'rectangular', 'hexagonal'}) – The grid type

Notes

The grid uses Cartesian indexing (indexing='xy' for NumPy’s meshgrid function). This implies that the grid’s shape will be (number of y coordinates) x (number of x coordinates).
If a range is zero, the step size is irrelevant.

Examples

You can iterate through a grid as if it were a list:

>>> grid = Grid2D((0, 1), (2, 3))
>>> for x, y in grid:
...     print(x, y)
0.0 2.0
1.0 2.0
0.0 3.0
1.0 3.0

plot(transform=None, label=None, style='hull', autoscale=True, zorder=None, ax=None, figsize=None, fc='gray')[source]¶

Plot the extension of the grid

Parameters:

transform (function, optional) – A coordinate transform to be applied to the (x,y) coordinates of the grid (e.g., Curcio1990Transform.dva2ret). It must accept two input arguments (x and y) and output two variables (the transformed x and y).
label (str, optional) – A name to be used as the label of the matplotlib plot. This can be used to label plots with multiple regions (i.e. call plt.legend after)
style ({'hull', 'scatter', 'cell'}, optional) –
- ‘hull’: Show the convex hull of the grid (that is, the outline of the smallest convex set that contains all grid points).
- ’scatter’: Scatter plot all grid points
- ’cell’: Show the outline of each grid cell as a polygon. Note that this can be costly for a high-resolution grid.
autoscale (bool, optional) – Whether to adjust the x,y limits of the plot to fit the implant
zorder (int, optional) – The Matplotlib zorder at which to plot the grid
ax (matplotlib.axes._subplots.AxesSubplot, optional) – A Matplotlib axes object. If None, will either use the current axes (if exists) or create a new Axes object
figsize ((float, float), optional) – Desired (width, height) of the figure in inches
fc (str or valid matplotlib color, optional) – Facecolor, or edge color if style=scatter, of the plotted region Defaults to gray

pulse2percept.utils.parfor(func, in_list, out_shape=None, n_jobs=-1, engine=None, scheduler='threading', func_args=[], func_kwargs={})[source]¶

Parallel for loop for NumPy arrays

Parameters:

func (callable) – The function to apply to each item in the array. Must have the form: func(arr, idx, *args, *kwargs) where arr is an ndarray and idx is an index into that array (a tuple). The Return of func needs to be one item (e.g. float, int) per input item.
in_list (list) – All legitimate inputs to the function to operate over.
out_shape (int or tuple of ints, optional) – If set, output will be reshaped accordingly. The new shape should be compatible with the original shape. If an integer, then the result will be a 1-D array of that length. One shape dimension can be -1. In this case, the value is inferred from the length of the array and remaining dimensions.
n_jobs (integer, optional, default: 1) – The number of jobs to perform in parallel. -1 to use all cpus
engine (str, optional, default: JobLib or Dask (if available), else serial) – {‘dask’, ‘joblib’, ‘serial’} The last one is useful for debugging – runs the code without any parallelization.
scheduler (str, optional, default: 'threading') – Which scheduler to use (irrelevant for ‘serial’ engine): - ‘threading’: a scheduler backed by a thread pool - ‘multiprocessing’: a scheduler backed by a process pool
*func_args (list, optional) – Positional arguments to func
**func_kwargs (dict, optional) – Keyword arguments to func

Returns:

ndarray – NumPy array of identical shape to arr

Note

Equivalent to pyAFQ version (blob e20eaa0 from June 3, 2016): https://github.com/arokem/pyAFQ/blob/master/AFQ/utils/parallel.py

pulse2percept.utils.pol2cart(theta, rho)[source]¶

Convert polar to Cartesian coordinates

Parameters:	rho (theta,) – The polar coordinates
Returns:	x, y (scalar or array-like) – The transformed Cartesian coordinates

class pulse2percept.utils.PrettyPrint[source]¶

An abstract class that provides a way to prettyprint all class attributes, inspired by scikit-learn.

Classes deriving from PrettyPrint are required to implement a _pprint_params method that returns a dictionary containing all the attributes to prettyprint.

Examples

>>> from pulse2percept.utils import PrettyPrint
>>> class MyClass(PrettyPrint):
...     def __init__(self, a, b):
...         self.a = a
...         self.b = b
...
...     def _pprint_params(self):
...         return {'a': self.a, 'b': self.b}
>>> MyClass(1, 2)
MyClass(a=1, b=2)

pulse2percept.utils.r2_score(y_true, y_pred)[source]¶

Calculate R² (the coefficient of determination)

The r2_score function computes the coefficient of determination, usually denoted as R².

The best possible score is 1.0, lower values are worse.

It represents the proportion of variance (of y) that has been explained by the independent variables in the model. It provides an indication of goodness of fit and therefore a measure of how well unseen samples are likely to be predicted by the model.

If $\hat{y}_i$ is the predicted value of the $i$-th sample and $y_i$ is the corresponding true value for total $n$ samples, the estimated R² is defined as:

\[R^2(y, \hat{y}) = 1 - \frac{\sum_{i=1}^{n} (y_i - \hat{y}_i)^2}{\sum_{i=1}^{n} (y_i - \bar{y})^2}\]

where $\bar{y} = \frac{1}{n} \sum_{i=1}^{n} y_i$ and $\sum_{i=1}^{n} (y_i - \hat{y}_i)^2 = \sum_{i=1}^{n} \epsilon_i^2$.

Note that r2_score calculates unadjusted R² without correcting for bias in sample variance of y.

New in version 0.7.

Parameters:	y_true (array-like) – Ground truth (correct) target values. y_pred (array-like) – Estimated target values.
Returns:	z (float) – The R² score

Notes

If the ground-truth data has zero variance, R² will be zero.
This is not a symmetric function

pulse2percept.utils.sample(sequence, k=1)[source]¶

Randomly selects k elements from a sequence

New in version 0.8.

Parameters:	sequence (list, tuple, np.ndarray) – A sequence like a list, a tuple, an array, etc. k (int or float, optional) – If an integer, the number of elements to pick If a float between 0 and 1, the fraction of elements to pick
Returns:	sample (list) – List of randomly chosen elements from the sequence

pulse2percept.utils.scale_image(img, scaling_factor)[source]¶

Scale the image foreground

This function scales the image foreground by a factor. The background of the image is assumed to be black (0 grayscale).

New in version 0.7.

Parameters:	img (ndarray) – A 2D NumPy array representing a (height, width) grayscale image, or a 3D NumPy array representing a (height, width, channels) RGB image scaling_factor (float) – Factory by which to scale the image
Returns:	img (ndarray) – A copy of the scaled image

pulse2percept.utils.shift_image(img, shift_cols, shift_rows)[source]¶

Shift the image foreground

This function shifts the center of mass (CoM) of the image by the specified number of rows and columns. The background of the image is assumed to be black (0 grayscale).

New in version 0.7.

Parameters:	img (ndarray) – A 2D NumPy array representing a (height, width) grayscale image, or a 3D NumPy array representing a (height, width, channels) RGB image shift_cols (float) – Number of columns by which to shift the CoM. Positive: to the right, negative: to the left shift_rows (float) – Number of rows by which to shift the CoM. Positive: downward, negative: upward
Returns:	img (ndarray) – A copy of the shifted image

pulse2percept.utils.trim_image(img, tol=0, return_coords=False)[source]¶

Remove any black border around the image

New in version 0.7.

Parameters:

img (ndarray) – A 2D NumPy array representing a (height, width) grayscale image, or a 3D NumPy array representing a (height, width, channels) RGB image. If an alpha channel is present, the image will first be blended with black.
tol (float, optional) – Any pixels with gray levels > tol will be trimmed.
return_coords (bool, optional) – If True, will also return the row and column coordinates of the retained image

Returns:

img (ndarray) – A copy of the image with trimmed borders.
(row_start, row_end) (tuple, optional) – The range of row indices in the trimmed image (returned only if return_coords is True)
(col_start, col_end) (tuple, optional) – The range of column indices in the trimmed image (returned only if return_coords is True)

pulse2percept.utils.unique(a, tol=1e-06, return_index=False)[source]¶

Find the unique elements of a sorted 1D array

Special case of numpy.unique (array is flat, sortened) with a tolerance level tol.

New in version 0.7.

Parameters:

a (array_like) – Input array: must be sorted, and will be flattened if it is not already 1-D.
tol (float, optional) – If the difference between two elements in the array is smaller than tol, the two elements are considered equal.
return_index (bool, optional) – If True, also return the indices of a that result in the unique array.

Returns:

unique (ndarray) – The sorted unique values
unique_indices (ndarray, optional) – The indices of the first occurrences of the unique values in the original array. Only provided if return_index is True.

class pulse2percept.utils.VisualFieldMap[source]¶

Base class for a visual field map (retinotopy)

A template

dva2ret(x, y)[source]¶: Convert degrees of visual angle (dva) to retinal coords (um)

ret2dva(x, y)[source]¶: Convert retinal coords (um) to degrees of visual angle (dva)

class pulse2percept.utils.Watson2014DisplaceMap[source]¶

Converts between visual angle and retinal eccentricity using RGC: displacement [Watson2014]

Converts from eccentricity (defined as distance from a visual center) in degrees of visual angle (dva) to microns on the retina using Eqs. 5, A5, and A6 in [Watson2014].

In a central retinal zone, the retinal ganglion cell (RGC) bodies are displaced centrifugally some distance from the inner segments of the cones to which they are connected through the bipolar cells, and thus from their receptive field. The displacement function is described in Eq. 5 of [Watson2014].

dva2ret(xdva, ydva)[source]¶

Converts dva to retinal coords

Parameters:	ydva (xdva,) – x,y coordinates in dva
Returns:	xret, yret (double or array-like) – Corresponding x,y coordinates in microns

ret2dva(xret, yret)[source]¶

Converts retinal distances (um) to visual angles (deg)

This function converts an eccentricity measurement on the retinal surface(in micrometers), measured from the optic axis, into degrees of visual angle using Eq. A6 in [Watson2014].

Parameters:	y_um (x_um,) – Original x and y coordinates on the retina (microns) coords ({'cart', 'polar'}) – Whether to return the result in Cartesian or polar coordinates
Returns:	x_dva, y_dva (double or array-like) – Transformed x and y coordinates (degrees of visual angle, dva)

static watson_displacement(r, meridian='temporal')[source]¶

Ganglion cell displacement function

Implements the ganglion cell displacement function described in Eq. 5 of [Watson2014].

Parameters:

r (double|array-like) – Eccentricity in degrees of visual angle (dva)
meridian ('temporal' or 'nasal') –

Returns:

The displacement in dva experienced by ganglion cells at eccentricity
r.

class pulse2percept.utils.Watson2014Map[source]¶

Converts between visual angle and retinal eccentricity [Watson2014]

static dva2ret(x_deg, y_deg, coords='cart')[source]¶

Converts visual angles (deg) into retinal distances (um)

This function converts degrees of visual angle into a retinal distance from the optic axis (um) using Eq. A5 in [Watson2014].

Parameters:	y_dva (x_dva,) – Original x and y coordinates (degrees of visual angle, dva) coords ({'cart', 'polar'}) – Whether to return the result in Cartesian or polar coordinates
Returns:	x_ret, y_ret (double or array-like) – Transformed x and y coordinates on the retina (microns)

static ret2dva(x_um, y_um, coords='cart')[source]¶

Converts retinal distances (um) to visual angles (deg)

This function converts an eccentricity measurement on the retinal surface(in micrometers), measured from the optic axis, into degrees of visual angle using Eq. A6 in [Watson2014].

Parameters:	y_um (x_um,) – Original x and y coordinates on the retina (microns) coords ({'cart', 'polar'}) – Whether to return the result in Cartesian or polar coordinates
Returns:	x_dva, y_dva (double or array-like) – Transformed x and y coordinates (degrees of visual angle, dva)