Thursday, March 14, 2024

Intrinsic Representation in Geometric Learning

Target audience: Intermediate
Estimated reading time: 7'
Newsletter: Geometric Learning in Python          

Tensors and vectors, which depend on coordinate systems, can be perplexing even for seasoned practitioners. The third installment of our Geometric Learning in Python series addresses intrinsic and extrinsic coordinate systems, including a transformation from cartesian to polar coordinates.

Table of contents
Introduction
        Contravariant vectors
        Covariant vectors
Follow me on LinkedIn


What you will learn: How to apply coordinates system and leverage intrinsic and extrinsic geometry.

Notes:

Introduction

In geometry, a coordinate system employs one or more numerical values, known as coordinates, to uniquely identify the location of points or other geometric entities within a manifold, like Euclidean space.

This article is the 3rd installment in our series on Geometric Learning in Python following
  1. Foundation of Geometric Learning introduces differential geometry as a applied to machine learning and its basic components.
  2. Differentiable Manifolds for Geometric Learning describes manifold components such as tangent vectors, geodesics with implementation in Python for Hypersphere using the Geomstats library.

This article does not offer a formal explanation of vectors, tensors, or coordinate systems. Readers are strongly recommended to seek out tutorials [ref 1], video series [ref 2], or publications [ref 3, 4] for more detailed information.

Note: Classes ManifoldPoint and HypersphereSpace have been introduced in the previous article [ref 5]

Covariance and contravariance

Contravariant and covariant vectors (or tensors) are fundamental concepts in the field of tensor calculus and differential geometry, used to describe geometric and physical entities in a way that is independent of the coordinate system used.

Note: This section provides a summary of the essential aspects of covariant and contravariant vectors. Readers are highly encouraged to refer to relevant tutorials [ref 6] for a more comprehensive understanding.

Contravariant vectors

Contravariant vectors, also known as tangent vectors, are vectors that transform in the same way as the coordinate system's axes. When the coordinate system is stretched or compressed, the components of a contravariant vector change in the same direction.
If the coordinate system is changed, the components of a contravariant vector transform inversely to the coordinates themselves.
A contravariant vector v is defined by its components vi and its basis vectors e:
\[\vec{v}=\sum_{i=1}^{n} v^{i} \vec{e}_{i}\]Contravariant vectors can be thought of as "pointing" in a certain direction in space, defining directions along which scalar quantities increase.

Covariant vectors

Covariant vectors, or dual vectors, are vectors whose components transform oppositely to the coordinate system's transformation. This means when the coordinate system is stretched, the components of a covariant vector are compressed, and vice versa.
The components of a covariant vector transform in the same way as the coordinate differentials, meaning they scale directly with changes in the coordinate grid.

Covariant vectors are often interpreted as describing surfaces or gradients that interact with other vectors to produce scalar quantities, like the dot product.\[\begin{matrix} \omega : \vec{v} \rightarrow < \ ,\vec{v}>\\ \omega(\vec{v} ) = <\vec{\omega}, \vec{v} > = \omega.v^{T} \end{matrix}\]

Intrinsic vs extrinsic geometries

Intrinsic coordinates

Intrinsic geometry involves studying objects, such as vectors, based on coordinates (or base vectors) intrinsic to the manifold's point. For example, analyzing a two-dimensional vector on a three-dimensional sphere using the sphere's own coordinates.
Intrinsic geometry is crucial in differential geometry because it allows the study of geometric objects as entities in their own right, independent of the surrounding space. This concept is pivotal in understanding complex geometric structures, such as those encountered in general relativity and manifold theory, where the internal geometry defines the fundamental properties of the space.

Here are key points:
  • Internal Measurements: Intrinsic geometry involves measurements like distances, angles, and curvatures that are internal to the manifold or surface. 
  • Coordinate Independence: Intrinsic coordinates describe points on a geometric object using a coordinate system that is defined on the object itself, rather than in the ambient space. 
  • Curvature: A fundamental concept in intrinsic geometry is curvature, which describes how a geometric object bends or deviates from being flat. The curvature is intrinsic if it is defined purely by the geometry of the object and not by how it is embedded in a higher-dimensional space.
Fig 1. Visualization of intrinsic coordinates of manifold in 3-dimension Euclidean space


Extrinsic coordinates

Extrinsic geometry studies objects relative to the ambient Euclidean space in which the manifold is situated, such as viewing a vector on a sphere's surface in three-dimensional space.

Extrinsic geometry in differential geometry deals with the properties of a geometric object that depend on its specific positioning and orientation in a higher-dimensional space. It is concerned with how the object is embedded in this surrounding space and how it interacts with the ambient geometry. It contrasts with intrinsic geometry, where the focus is on properties that are independent of the ambient space, considering only the geometry contained within the object itself. Extrinsic analysis is crucial for understanding how geometric objects behave and interact in larger spaces, such as in the study of embeddings, gravitational fields in physics, or complex shapes in computer graphics.

Here’s what extrinsic geometry involves:
  • Embedding: Extrinsic geometry examines how a lower-dimensional object, like a surface or curve, is situated within a higher-dimensional space. 
  • Extrinsic Coordinates: These are coordinates that describe the position of points on the geometric object in the context of the larger space.
  • Normal Vectors: An important concept in extrinsic geometry is the normal vector, which is perpendicular to the surface at a given point. These vectors help in defining and analyzing the object's orientation and curvature relative to the surrounding space.
  • Curvature: Extrinsic curvature measures how a surface bends within the larger space. 
  • Projection and Shadow: Extrinsic geometry also looks at how the object projects onto other surfaces or how its shadow appears in the ambient space.
Fig. 2 Visualization of extrinsic in 3-dimension Euclidean space


Implementation

Manifold and coordinates system

We will illustrate the various coordinates on the hypersphere space we introduced in a previous article Geometric Learning in Python: Manifolds
The structure of data points on a manifold is defined by the class ManifoldPoint described in a previous post ManifoldPoint definition. 

@dataclass
class ManifoldPoint:
    id: AnyStr
    location: np.array
    tgt_vector: Optional[List[float]] = None
    geodesic: Optional[bool] = False
    intrinsic: Optional[bool] = False


   def to_intrinsic(self, space: LevelSet) -> np.array:
       return space.extrinsic_to_intrinsic_coords(self.location) if not self.intrinsic else self.location
 

   def to_extrinsic(self, space: LevelSet) -> np.array: 
       return space.intrinsic_to_extrinsic_coords(self.location) if self.intrinsic else self.location

   def to_intrinsic_polar(self, space: LevelSet) -> np.array:
       intrinsic_coordinates = self.to_intrinsic(space)
       return ManifoldPoint.__cartesian_to_polar(intrinsic_coordinates)

The class ManifoldPoint has 3 methods to define its location given a coordinate system:
  • to_intrinsic: Convert the current location from extrinsic cartesian coordinates (3 dimension) to intrinsic cartesian coordinates (2 dimension) if the flag intrinsic is False
  • to_extrinsic: Convert the location from intrinsic cartesian coordinates to extrinsic coordinates if the flag intrinsic is True
  • to_intrinsic_polar: Convert the current location from extrinsic cartesian coordinates (3 dimension) to intrinsic polar coordinates (2 dimension) if the flag intrinsic is False
This last method relies on the transformation from Cartesian to Polar coordinate on the tangent plane.

The following figure illustrates the transformation from cartesian coordinates (x, y) to polar coordinates (r, theta)


Fig. 3 Visualization of polar coordinates on 2 dimension surface

Here are the mathematical equations for the transformation from cartesian to polar coordinates.\[\begin{matrix} r=\sqrt{x^{2}+y^{2}}\ \ \ \ x.y > 0\\ \theta = arccos(\frac{x}{r})\ \ if\ y >= 0 \ \ \ -arccos(\frac{x}{r}))\ \ \ if\ y <0 \end{matrix}\]
The following  private static method  __cartesian_to_polar , which executes the two formulas, is straightforward.

def __cartesian_to_polar(cartesian_coordinates: np.array) -> np.array:
   x = cartesian_coordinates[0]
   y = cartesian_coordinates[1]


   # Compute rho, theta from x, y 
  r = math.sqrt(x ** 2 + y ** 2)          // First equation
  theta = math.acos(x / r) if y >= 0.0 else -math.acos(x / r)  // Second equation
 
  return np.array([r, theta])

Next, we will explore how to calculate the mean of multiple points on a manifold, both in Euclidean space and along the geodesic of the manifold in extrinsic coordinates.

Euclidean and Fréchet means

Let's explore how to compute means from the perspective of the extrinsic geometry considering a surface that is embedded in a three-dimensional Euclidean space.


Fig. 4 Visualization of Euclidean mean (extrinsic) and Fréchet mean (intrinsic)


First let's compute the mean of multiple point in the default Euclidean space using comprehensive list and Numpy mean functions

def euclidean_mean(manifold_points: List[ManifoldPoint]) -> np.array:
    return np.mean([manifold_pt.location for manifold_pt in manifold_points], axis=0)

Next we compute the intrinsic mean or Fréchet mean using Geomstats class FrechetMean.

def frechet_mean(self, manifold_pt1: ManifoldPoint, manifold_pt2: ManifoldPoint) -> np.array:
   from geomstats.learning.frechet_mean import FrechetMean


   frechet_mean = FrechetMean(self.space) 
   x = np.stack((manifold_pt1.location, manifold_pt2.location), axis=0) 
   frechet_mean.fit(x)
 

   return frechet_mean.estimate_


Similar to the prior article on Manifolds [ref 7] in this series, we utilize the Hypersphere as our evaluation manifold. In the ensuing implementation, we generate two random points on a Hypersphere, calculate the Euclidean and Fréchet means, and then depict the outcomes using the Matplotlib library.

The execution confirms that euclidean mean does not belong to the manifold (as with fig 4) while the Fréchet mean do.

Note: The HypersphereSpace class, introduced in the previous article [ref 7], encapsulates the essential methods for manipulating the Hypersphere function defined in the Geomstats library.

manifold = HypersphereSpace(True)
samples = manifold.sample(2)
assert(manifold.belongs(samples[0]))   # Is True
        
# 1. Generate points on the Hyperphere
vector = [0.8, 0.4, 0.7]
manifold_points = [
ManifoldPoint(id=f'data{index}', location=sample, tgt_vector=vector, geodesic=False) 
   for index, sample in enumerate(samples)
]
        
# 2. Compute mean of two points on the Euclidean space
euclidean_mean = manifold.euclidean_mean(manifold_points)
assert(manifold.belongs(euclidean_mean))   # Is False
        
# 3. Compute the mean on the manifold geodesics
frechet_mean = manifold.frechet_mean(manifold_points[0], manifold_points[1])
print(f'Euclidean mean: {euclidean_mean}\nFrechet mean: {frechet_mean}')
assert (manifold.belongs(frechet_mean))   # Is True


# 4. Visualize various data points on the Hypersphere
frechet_pt = ManifoldPoint(id='Frechet mean',location=frechet_mean, tgt_vector=None,geodesic=False)
manifold_points.append(frechet_pt)
manifold.show_manifold(manifold_points, [euclidean_mean])


Output:
Euclidean mean: [ 0.52305713,  -0.39238391,  -0.10012601]
Fréchet mean: [ 0.79071656, -0.59317508, -0.15136262]

The method show_manifold [ref 8] is called to display the two original points, data0 and data1, on the hypersphere, along with the Fréchet and Euclidean means. It becomes clear that the Euclidean mean does not reside on the hypersphere.
Fig. 5 Illustration of Fréchet and Euclidean mean on a Hypersphere


Intrinsic cartesian coordinates

In the previous article, we explored tangent space and tangent vectors [ref 9] for the hypersphere using extrinsic coordinates. The figure below demonstrates the tangent space, showing both the Euclidean cartesian coordinates X, Y, Z, and the intrinsic cartesian coordinates x, y.
Fig. 6 Visualization of intrinsic cartesian coordinates on a hypersphere


The method extrinsic_to_intrinsic transforms the position of a list of points on the hypersphere from extrinsic cartesian coordinates to intrinsic coordinates. Conversely, the method intrinsic_to_extrinsic changes their location from intrinsic cartesian coordinates to extrinsic coordinates.

def extrinsic_to_intrinsic(self, manifold_pts: List[ManifoldPoint]) -> List[ManifoldPoint]:
   return [ManifoldPoint(
         id=pt.id,
         location=pt.to_intrinsic(self.space),
         tgt_vector=pt.tgt_vector,
         geodesic=pt.geodesic,
         intrinsic=True) for pt in manifold_pts]



def intrinsic_to_extrinsic(self, manifold_pts: List[ManifoldPoint]) -> List[ManifoldPoint]:
    return [ManifoldPoint(
        id=pt.id,
        location=pt.to_extrinsic(self.space),
        tgt_vector=pt.tgt_vector,
        geodesic=pt.geodesic,
        intrinsic=False) for pt in manifold_pts]


The code snippet below transforms two randomly chosen points on the hypersphere from extrinsic coordinates (3-dimensional) to intrinsic coordinates (2-dimensional) and then back to extrinsic coordinates.

intrinsic = False
manifold = HypersphereSpace(True, intrinsic)
        
# Create Manifold points with default extrinsic coordinates
random_samples = manifold.sample(2)
manifold_pts = [
  ManifoldPoint(f'id{index}', value, None, False, intrinsic) 
  for index, value in enumerate(random_samples)
]


print(f'Extrinsic coordinates:\n{[m_pt.location for m_pt in manifold_pts]}')
intrinsic_pts = manifold.extrinsic_to_intrinsic(manifold_pts)
print(f'Intrinsic coordinates:\n{[m_pt.location for m_pt in intrinsic_pts]}' )

extrinsic_manifold_pts = manifold.intrinsic_to_extrinsic(intrinsic_manifold_pts)
print(f'Regenerated extrinsic coordinates:\n{[m_pt.location for m_pt in extrinsic_manifold_pts]}')


Output:
Extrinsic coordinates:
[[-0.53207445, -0.68768687,  0.49394691], 
 [-0.45305036, -0.1875893 , -0.87152489]]
Intrinsic Coordinates:
[[1.05417307, 4.05388149], 
 [2.62909989, 3.53415928)]
Regenerated extrinsic coordinates:
[[-0.53207445, -0.68768687,  0.49394691], 
[-0.45305036, -0.1875893 , -0.87152489]]

As expected the final output of the coordinates of the two points matches the original extrinsic coordinates.

Intrinsic polar coordinates

The figure below demonstrates the tangent space, showing both the Euclidean cartesian coordinates X, Y, Z, and the intrinsic cartesian coordinates r, theta.
Fig. 7 Visualization of intrinsic polar coordinates on a hypersphere

The extrinsic_to_intrinsic_polar method functions similarly to extrinsic_to_intrinsic for cartesian coordinates. It changes the location of a set of points on the hypersphere from extrinsic cartesian coordinates to intrinsic polar coordinates.

def extrinsic_to_intrinsic_polar(self, manifold_pts: List[ManifoldPoint]) -> List[ManifoldPoint]:
  return [ManifoldPoint(
      id=pt.id,
      location=pt.to_intrinsic_polar(self.space),
      tgt_vector=pt.tgt_vector,
      geodesic=pt.geodesic,
      intrinsic=False) for pt in manifold_pts]

Output:
To extrinsic: [0.25934338  0.14167993  0.95533649]
To intrinsic: [2.09057978  3.26478145]
To polar: [3.16428762  0.64002724]

References

[3] Vector and Tensor Analysis with Applications -  A. I. Borisenko, I. E. Tarapov - Dover Books on MathenaticsPublications 1979 
[4] A Student's Guide to Vectors and Tensors - D. Fleisch - Cambridge University Press - 2008


-------------
Patrick Nicolas has over 25 years of experience in software and data engineering, architecture design and end-to-end deployment and support with extensive knowledge in machine learning. 
He has been director of data engineering at Aideo Technologies since 2017 and he is the author of "Scala for Machine Learning", Packt Publishing ISBN 978-1-78712-238-3 and Geometric Learning in Python Newsletter on LinkedIn.
Posted by No comments:
Labels: , , , , , , , , ,

Saturday, February 17, 2024

Geometric Learning in Python: Differential Operators

Target audience: Intermediate
Estimated reading time: 5'
The use of Riemannian manifolds and topological computing is gaining traction in machine learning circles as a means to surpass the existing constraints of data analysis within Euclidean space. 
This article aims to present the fundamental aspects of differential geometry.


Table of content
      Why SymPy
      Gradient
      Divergence
      Curl
      Validation
      Laplace
      Fourier
Follow me on LinkedIn

What you will learn: Vector fields, Differential operators and integral transforms as key components of differential geometry applied to deep learning.

Notes: 
  • Environments: Python  3.10.10 and SymPy 1.12
  • The differential operators are used in a future post dedicated to differential geometry and manifold learning
  • This article assumes that the reader is familiar with differential calculus and basic concept of geometry
  • Source code is available at github.com/patnicolas/Data_Exploration/diffgeometry
  • To enhance the readability of the algorithm implementations, we have omitted non-essential code elements like error checking, comments, exceptions, validation of class and method arguments, scoping qualifiers, and import statements.

Overview

The heavy dependence on elements from Euclidean space for data analysis and pattern recognition in machine learning might be approaching a threshold. Incorporating differential geometry and Riemannian manifolds has already shown beneficial effects on the accuracy of deep learning models, as well as on reducing their development costs [ref 1].

Why differential geometry

Conventional approaches in machine learning, pattern recognition, and data analysis often presuppose that input data can be effectively represented using elements from Euclidean space. Although this assumption has been sufficient for numerous applications in the past, there's a growing awareness among data scientists and engineers that most data in vision and pattern recognition actually reside in a differential manifold (feature space) that's embedded within the raw data's Euclidean space. 
Leveraging this geometric information can result in a more precise representation of the data's inherent structure, leading to improved algorithms and enhanced performance in real-world applications [ref 2].

Why SymPy

SymPy is a Python library dedicated to symbolic mathematics. Its implementation is as simple as possible in order to be comprehensible and easily extensible. It is licensed under BSD. SymPy supports differential and integral calculus, Matrix operations, algebraic and polynomial equations, differential geometry, probability distributions, 3D plotting and discrete math [ref 3].
To install pip install sympy
Source code for SymPy is available at github.com/sympy/sympy.git

This article has 3 sections:
  1. Introduction of vector fields and implementation in SymPy
  2. Differential operators (Gradient, Divergence and Curl)
  3. Laplace and Fourier Transform

Vector fields

Let's start with an overview of vector fields as the basis of differential geometry [ref 4].
Consider a three-dimensional Euclidean space characterized by basis vectors {ei} and a vector field V, expressed as (f1, f2, f3). In this scenario, it is crucial to note that we are following the conventions of Einstein's tensor notation.

Visualization of a vector field in 3D Euclidean space

The vector field at the point P(x,y,z) is defined as the tuple (f1(x,y,z), f2(x,y,z), f3(x,y,z)). The vector over a field of n dimension field and basis vectors  {ei}  can be formally defined as \[f: \boldsymbol{x} \,\, \epsilon \,\,\, \mathbb{R}^{n} \mapsto \mathbb{R} \\ f(\mathbf{x})=\sum_{i=1}^{n}{f^{i}}(\mathbf{x}).\mathbf{e}_{i}\] Example for 3 dimension space: \[f(x,y,z) = (2x+z^{3})\boldsymbol{\mathbf{\overrightarrow{i}}} + (xy+e^{-y}-z^{2})\boldsymbol{\mathbf{\overrightarrow{j}}} + \frac{x^{3}}{y}\boldsymbol{\mathbf{\overrightarrow{k}}}\] Let's carry out the creation of this vector field, f, in SymPy and compute its value at f(1.0, 2.0, 0.5). To begin, we need to establish a 3D coordinate system, denoted as r, and then express the vector using r.

from sympy.vector import CoordSys3D
from sympy import exp


r = CoordSys3D('r')
f = (2*r.x + r.z**3)*r.i + (r.x*r.y+sympy.exp(-r.y)-r.z**2)*r.j + (r.x**3/r.y)*r.k

w = f.evalf(subs={r.x: 1.0, r.y: 2.0, r.z: 0.2})
print(w)                                 # 2.008*r.i + 2.09533528323661*r.j + 0.5*r.k

Now, let's consider the same vector V with a second reference; origin 0' and basis vector e'i
Visualization of two coordinate systems

\[f(\mathbf{x})=\sum_{i=1}^{n} f'^{i}(\mathbf{x}).\mathbf{e'}_{i}\] The transformation matrix Sij convert the coordinates value functions  fi and f'i. The tuple f =(fi) is the co-vector field for the vector field V 
\[S_{ij}: \begin{Vmatrix} f^{1} \\ f^{2} \\ f^{3} \end{Vmatrix} \mapsto \begin{Vmatrix} f'^{1} \\ f'^{2} \\ f'^{3} \end{Vmatrix}\] The scalar product of the co-vector f' and vector v(f) is defined as \[< f',v> = \sum f'_{i}.f^{i}\] Given the scalar product we can define the co-vector field f' as a linear map \[\alpha (v) = < f',v> (1) \] 

Differential operators

Let's demonstrate how to calculate differential operators using the SymPy library within a 3D Cartesian coordinate system. To begin, we'll create a class named DiffOperators which will encapsulate the implementation of various differential operators and transforms.

Gradient

Consider a scalar field f in a 3-dimension space. The gradient of this field is defined as the vector of the 3 partial derivatives f with respect to x, y and z [ref 5].\[\triangledown f= \frac{\partial f}{\partial x} \vec{i} + \frac{\partial f}{\partial y} \vec{j} + \frac{\partial f}{\partial z} \vec{k}\]

class DiffOperators(object):
    def __init__(self, expr: Expr):
        self.expr = expr

    def gradient(self) -> VectorZero:
        from sympy.vector import gradient

        return gradient(self.expr, doit=True)
Example of input function: \[f(x,y,z)=x^{2}yz\]
r = CoordSys3D('r')
this_expr = r.x*r.x*r.y*r.z

diff_operator = DiffOperators(this_expr)
grad_res = diff_operator.gradient()
\[\triangledown f(x,y,z) = 2xyz.\vec{i} + x^{2}z.\vec{j} + x^{2}y.\vec{k}\]

Divergence

Divergence is a vector operator used to quantify the strength of a vector field's source or sink at a specific point, producing a signed scalar value. When applied to a vector F, comprising components X, Y, and Z, the divergence operator consistently yields a scalar result [ref 6].
\[div(F)=\triangledown .F=\frac{\partial X}{\partial x}+\frac{\partial Y}{\partial y}+\frac{\partial Z}{\partial z}\]
def divergence(self, base_vec: Expr) -> VectorZero:
    from sympy.vector import divergence

    div_vec = self.expr*base_vec
    return divergence(div_vec, doit=True)
Using the same input expression as used for the gradient calculation:
div_res = diff_operator.divergence(r.i + r.j + r.k)
\[div(f(x,y,z)[\vec{i} + \vec{j}+ \vec{k}]) = 2x(y+z+xy)\]

Curl

In mathematics, the curl operator represents the minute rotational movement of a vector in three-dimensional space. This rotation's direction follows the right-hand rule (aligned with the axis of rotation), while its magnitude is defined by the extent of the rotation [ref 6]. Within a 3D Cartesian system, for a three-dimensional vector F, the curl operator is defined as follows:
\[ \triangledown * \mathbf{F}=\left (\frac{\partial F_{z}}{\partial y}- \frac{\partial F_{y}}{\partial z} \right ).\vec{i} + \left (\frac{\partial F_{x}}{\partial z}- \frac{\partial F_{z}}{\partial x} \right ).\vec{j} + \left (\frac{\partial F_{y}}{\partial x}- \frac{\partial F_{x}}{\partial y} \right ).\vec{k} \]
def curl(self, base_vectors: Expr) -> VectorZero:
    from sympy.vector import curl

    curl_vec = self.expr*base_vectors
    return curl(curl_vec, doit=True)

If we use only the two base vectors j and k:
curl_res = diff_operator.curl(r.j+r.k)
\[curl(f(x,y,z)[\vec{j} + \vec{k}]) = x^{2}\left ( z-y \right ).\vec{i}-2xyz.\vec{i}+2xyz.\vec{k}\]

Validation

We can confirm the accuracy of the gradient, divergence, and curl operators implemented in SymPy by utilizing the subsequent formulas:  \[curl(\triangledown f))= \triangledown *\left ( \triangledown f\right ) = 0\]  and  \[\triangledown . curl (F) = \triangledown .\left ( \triangledown * F \right ) = 0\]
grad_res = diff_operator.gradient()
assert(diff_operator.curl(grad_res) ==0)   # Print 0

curl_res = diff_operator.curl(r.i + r.j + r.k)
assert(diff_operator.divergence(curl_res) == 0) # Print 0


Transforms

An integral transform transforms a function from the time domain into its equivalent representation in the frequency domain, as depicted in the following diagram:

Illustration of conversion from time domain to frequency domains

The evaluation of SymPy functions follows a two-step process:
  1. Generate function F in frequency space
  2. Evaluate output of F for given input values.

Laplace

The Laplace transform is a type of integral transform that changes a function of a real variable (typically time, t) into a function in the s-plane, which is a variable in the complex frequency domain [ref 7].. Specifically, it transforms ordinary differential equations into algebraic ones and converts convolution operations into simpler multiplication. The formulation of the Laplace transform is:\[\mathfrak{L}[f(t)](s)=\int_{0}^{+\inf}f(t).e^{-st}dt\]
def laplace(self):
    from sympy import laplace_transform
        
    return laplace_transform(self.expr, t, s, noconds=True)


Let's create the Laplace transforms for two basic functions and then calculate the outcomes of these transformed functions in the frequency domain, using the values s=0.5 and a=0.2.
t, s = sympy.symbols('t, s', real=True)
a = sympy.symbols('a', real=True)
diff_operator = DiffOperators(sympy.exp(-a*t))
laplace_func = diff_operator.laplace()


print(Laplace_func.evalf(subs={s:0.5, a:0.2}))   #  1.4285
\[\int_{0}^{+\inf}e^{-at}.e^{-st}dt = \frac{1}{a+s}\]
diff_operator = DiffOperators(sympy.sqrt(-a*t))
laplace_func = diff_operator.laplace() 

print(Laplace_func.evalf(subs={s:0.5, a:0.2}))   #  1.1209
\[\int_{0}^{+\inf}\sqrt{at}.e^{-st}dt = \frac{\sqrt{a\pi}}{2.s^{3/2}}\]

Fourier

Like the Laplace transform, the Fourier transform is another integral transform that translates a function into a version that represents the frequencies contained in the original function. The result of this transformation is a complex-valued function dependent on frequency [ref 8]. Often referred to as the frequency domain representation of the original function, the formulation of the Fourier transform is as follows: \[F(x )=\int_{-\inf}^{+\inf}f(t).e^{-2\pi itx}dt\] The implementation uses the SymPy function fourier_transform which takes the function (self.expr) and x, and k parameters as arguments.
def fourier(self):
   from sympy import fourier_transform
     
   return fourier_transform(self.expr, x, k)

Let's implement the Fourier transforms for two simple functions, exp and sin, and then calculate the outcomes of these transformed functions in the frequency domain with k=0.4
x, k = sympy.symbols('x, k', real=True)
diff_operator = DiffOperators(sympy.exp(-x**2))

fourier_func = diff_operator.fourier()

print(fourier_func.evalf(subs={k: 0.4}))       # 0.3654
\[\int_{-\inf}^{+\inf}e^{-t^{2}}.e^{-2\pi itx}dt = \sqrt{\pi}.e^{-2\pi x^{2}}\]
diff_operator = DiffOperators(sympy.sin(x))
diff_operator.fourier()
\[\int_{-\inf}^{+\inf}\sin(t).e^{-2\pi itx}dt = 0\]


Thank you for reading this article. For more information ...

References




---------------------------
Patrick Nicolas has over 25 years of experience in software and data engineering, architecture design and end-to-end deployment and support with extensive knowledge in machine learning. 
He has been director of data engineering at Aideo Technologies since 2017 and he is the author of "Scala for Machine Learning" Packt Publishing ISBN 978-1-78712-238-3










Posted by No comments:
Labels: , , , , , , , , , ,
Subscribe to: Posts (Atom)