Auxiliary Functions

Download this as a Jupyter notebook

This notebook covers the usage of the auxiliary function provided in pyXla.

A key feature of the pyXla framework is the separation of sampling and analysis. A set of functions are provided to support sampling. They are:

  1. pyxla.util.sample_X()

  2. pyxla.util.compute_F()

  3. pyxla.util.compute_V()

  4. pyxla.util.compute_D()

  5. pyxla.util.compute_N()

Each function corresponds to a given input file as indicated by it suffix.

When a sample is loaded declaratively via domain and function specification (method 3 in Loading and Sampling), these function are used under the hood. These function are available to the user for finer control over sampling.

The auxiliary functions are all imported from pyxla.util:

from pyxla.util import sample_X, compute_F, compute_V, compute_D, compute_N

One can generate an X file as below:

from pyxla.sampling import HilbertCurveSampler

sampler = HilbertCurveSampler(
    sample_size=100, dim=2, return_neighbourhood=True # will return an N file too
)

X, N = sample_X(sampler)
X.head()
x0 x1
0 11.086759 13.051422
1 5.908445 11.948770
2 1.000000 18.949877
3 17.168543 13.882606
4 16.745878 12.681372

Specifying return_neighbourhood=True generates an N file as well:

N.head()
id1 id2
0 0 1
1 1 2
2 2 3
3 3 4
4 4 5

The F file can be generated from the X file by specifying an objective function or multiple objective functions:

def sphere(X):
    return X[0]**2 + X[1]**2

def summation(X):
    return X.sum()

F = compute_F([sphere, summation], X)
F.head()
f0 f1
0 293.255841 24.138181
1 177.682813 17.857214
2 360.097846 19.949877
3 487.485613 31.051149
4 441.241633 29.427250

The process is similar for the V file:

V = compute_V([lambda X: sphere(X) - 2], X)
V.head()
v0
0 291.255841
1 175.682813
2 358.097846
3 485.485613
4 439.241633

Computing a D file is straightforward:

D = compute_D({"X": X}, metric='canberra') # you can define a function or specify any of scipy's distance functions
D.head()
d
id1 id2
0 1 0.348798
2 1.018849
3 0.246104
4 0.217707
5 1.326523

The N file is equally straightforward; either specify a neighbourhood function or supply the a literal: "hilbert-curve" or "X-index".

"hilbert-curve" uses the Hilbert curve to efficiently generate neighbourhood information (see pyxla.sampling.hilbert_curve_neighbour_sampling()).

"X-index" extracts neighbourhood from the default index of a DataFrame.

def randomly_neighbours(a, b):
    import random
    return random.choice([True, False])

N = compute_N({"X": X}, neighbourhood_func=randomly_neighbours)
N.head()
id1 id2
0 0 2
1 0 4
2 0 5
3 0 9
4 0 10