Chapter 11 Tools for Scientific Computing The purpose of this book is to prepare you to work in data science, the natural sciences and. engineering. We have addressed programming concepts and the Python language. In this chapter we will examine some of the tools used in scientific and engineering computations, 11.1 Numpy ‘Assume that you have collected temperature data for a project and that you have 10,000 collected data points. It would be absurd to write out 10,000 different variables to hold the values: [a,b,c % aa, ab, aC, »]- In this case using just lower case, you would need to go up to three letter variables (aaa, aab, aac, ..). There is not a concern about running out of variable names. The concer is that there is not a good way to access all of those variable names. It cannot be easily accessed in a loop. What we want is something like the vector idea found in math courses where you have 1/82, %3, 10000. Most programming languages support this through a data structure called an array. These languages extend the basic datatype to an array version. Mean- ing you can have arrays of integers, floats and other more exotic data containers. These arrays must have a uniform data type for the entire array. Python does not include a ba- sic array structure. It does have a more flexible structure known as a list will allows for elements of different data types. Lists were covered in an earlier chapter. To get a traditional array in Python, we use the NumPy module which is a collection of objects used to create and manipulate numerical data. NumPy provides an extension pack- age to Python for multi-dimensional arrays. Numpy is written in C and often optimised for special hardware and so core Numpy routines are at least an order of magnitude faster. ‘The design of the package reflects an array oriented computing view and is designed for scientific computing. Numpy is very fast and very efficient. Although Python can be rather slow, Numpy can achieve speeds associated with complied languages like C. Information and download of Numpy are at: https: //numpy.org/ . Numpy is part of the Anaconda distribution ‘We begin with creating a simple array of 32 bit integers.134 Chapter 11. Tools for Scientific Computing Sos a= mp.array((o, 1, 2, 31) prints) wis Note that when printing the array it looks like a list, but without the commas separating the values ‘We can get information about the datatype through [print (expe) eine (a. type) aeypeC' ints) An example of a two dimensional array [o-eperyti, 20% aD postage a> Numpy arrays are zero indexed. This means that the index for the array starts with 0 and not 1. Access is via square brackets as show here fa = ap-array((a7, 39, 23, 201) prine ala 2 We typically work with integer or floating point arrays. Ifall the elements are integers then that will be the datatype for the whole array, otherwise we get an array of floats: [a= ap amayttr, #23, to Pe parraysr, 39, 28, 29.00) In practice we don’t hand enter the data. Iti either read in froma file (see chapter 10) or it is created by an array generation function. Some very useful functions to create Numpy arrays follow. + To create a range of numbers starting at 4, up to but not including # with a separation of step, np.arange(a,b,step) * To create a range of N numbers starting at a, ending at b, np. linspace(a,b, N) + To create an array of zeros with shape (n,m), np.zeros ((n,n))11.1. Numpy 135 + To create an array of ones with shape (1,m), np-ones((n,m)) * To create an array of random numbers between 0 and 1 with shape (1m), np.random.rand(n,m) Please experiment to get a feel for Numpy arrays. print ep. Linapace 0,5, 500) We saw above that one can acces used to assign values array elements using the square brackets. This can be = ap zeros GSD) eta = 812 prince) Lo. 0, 612. 0, 0.) Numpy has many feature for array access and manipulation. Accessing parts of the array are known as slices. One such slice is the range operation: array[startiend:step] (begin at “start”, go up to but one less than “end” and use “step” as the stride): 3 > np.arangeCiO) prince) [0,4 2, 8, 4 5, 8, 7, 8 8) princes) B34 The slice is a view of the data - not a copy. ‘Fandine (0,55,15) # random array S) i print this out a= (3:6) # grap a alice 14 = ","a) # print out the alice 100 #’ et one of he elenents in the slice fied d=", a) # print d again to see the modification a) # print a to gen if it was changed 47, 10, 2, 80, 12, 6 34, 29, 43, 14, 44, 48, 28, 32, 40] 60, 32, 6) tied d * [100, 12, 6) = LAT, 10, 2, 100, 12, 6, 34, 29, 43, 14, 44, 48, 28, 32, 40) ‘To make a copy you should use the copy method so you don’t modify the original array: stil cony0 Arrays and loops go hand in hand. A for loop is an easy way to access or modify the elements of an array. For example, fill out an array of length 100 with 1/ (i+ 1) where jis the index. eee tor 4 in range(™) xt) = Ve)136 Chapter ‘Tools for Scientific Comp 11.2. Numpy operations Python is known to be slow. This partially has to do with the dynamic variables and the challenges of optimization in this environment. So, Numpy has a number of element-wise operations built-in. The basic binary operators are overloaded which means that when Python sees x +y it knows to call the element-wise addition function (under the hood), Automatic element-wise operations include * Addition of arrays: x+y + Addition with constants: x +10 + Scalar multiplication: ¢ x * Array multiplication: ab ‘+ Matrix multiplication (dot): np.dot (x,y) # Functions x*x+22x43,np-sin(e), np-exp(x) ‘These operations are very fast. This is where the power of Numpy starts to emerge. If you want to create a collection of points (x4, y,) where 0 < x¢ < 47, ye = af +e *sin(xy), with 0 < k < 200, the traditional element-wise approach would be CADsx(4) + methoxy (x (19) emer sinGela}) Using the overloaded operators in Numpy, we can rewrite this using the implicit element wise operations: ‘npore nunpy a= oP x = 200 5p. Lsnspace(0,enp.pi.8) x + mp expC amp sins) Because these arrays are objects, they have methods associated with them. ssa) femean() fxsta0) # standard Deviation -max0) mind Elementwise logical operations and comparisons can be done with Numpy arrays as well We will see more Numpy in a later section on plotting, 11.3. Random Values When doing simulations of natural systems, or testing code, or performing numerical op- timization, itis important to have access to random numbers. Python and Numpy provide values sampled from a variety of distributions. The generation of random values is a whole separate subject and one needs to keep in mind that the values produced by these routines11.3. Random Values 137 are not truly random and are not appropriate for cryptographic level randomness. How- ever, many applications work very well with pseudo-random sequences. Random and Numpy Random libraries both use the Mersenne twister sequence to generate their values and in most cases only one library is required, ‘The Random number library is accessed via import random and the Numpy ran- dom routines are brought in with import nunpy. The Random library is accessed via randon.function() and the Numpy random sample library is accessed via numpy random. function() A few routines in Random: + randint (low, high) - Produces a random integer where low ICT, RIC2, RIC3, RICH \a RCI [RIC | RICS | RICH RECT, R3C2, RIC3, RICA \n Assume the CSV files looks like 4,6.4,5.5,2.3,9.4 14.6,84.7,17.9,4.12 4,2,4,-4.8,66 0.88 1.2, 44 Recall to read a text file: Fin = open(iasca.cay') row = Tine.strip@) Prine (roe) tin.eloee) How would we get these values loaded into a Numpy two-dimensional array? Problem one is that a row is a single object, not a list of values. Second issue is that they come in as text. Tin = openvaaca.cay') row = Tine. strip) fSvens = Dine. sphie(*,") prine (itera) sin, etese0 Each line needs to be split and the data converted to floats. Then it can be loaded into a numpy array. If we know the size of the matrix, then itis fairly straightforward: fasta = npcenpty((9,5) type tin = open('aata.crv') tor Line in fin row = Line, seripQ) teas = ror.splat(’, cols = 0 detazovs cols] evanciten) If you don’t know the dimensions then a small modification will work:140. Chapter 11. Tools for Scientific Computing Jasta = np-array(() typ tin = opent'dava.cor') tor lane an tin row = Line, strip vens = row. split (*,") “t1est") ‘Or you can just do [trom nunpy import gente acta + gentzonext [from pandas import read 11.5 Matplotlib ‘There are a number of plotting options for Python. One of the oldest plotting packages and part of the SciPy collection is MatPlotLib. You can think of this as an array visualiza~ tion tool and not just a tool to produce graphs. This section will focus on traditional 2D plots with the goal of publication quality plots. Other applications such as 3D plots, ani- ‘mated plots, ete will not be addressed here. We have included references to the MatPlotLib documentation and references at the end, In order to use the routines you need to import the plotting package: For our first example, we will plot a simple data set: fe = mp array (10,1,2,3,49) fy = mp-areay((1.4,3.7,91) put .piet G9) [Pieisnev0 The resulting plot is shown in Figure 11.1(a). Itis that simple. The general idea is that you create the x and y arrays from your code (data analysis, simulation, etc). These are passed. into the plot command combined with any decorations you want. Matplotlib then renders the graphics. When using Jupyter Lab it attempts to render any of the plot commands, even a title command, plotting works differently than in the interpreter. So, you will normally place all of your plot commands related to a given plot in a single cell, Another example is given in Figure 11.1(b). ps Lanapace (0,1, 100) fy = xe(i-x) # Inverted parabola pit.plot(e,y, color = ‘green’, Linestyle pit-shovO dorted")