X/HEC Datascience for Business
2024/2025 Mathurin Massias
Lab no 1 : Introduction: Python, Numpy, Pandas
- Evaluation -
The Lab is done by pairs. There is no exception. On Moodle you have a section called “Lab
1 submission” for the first class. Each student in the pair must upload their notebook, with the
filename constructed as fistname1_LASTNAME1_firstname2_LASTNAME2.ipynb, where you have sub-
stituted your respective first and last names. Ex : mathurin_MASSIAS_sylvain_COMBETTES.ipynb.
There is no evaluation if you don’t respect this. If code is shared between groups, both groups
get 0. Read the intro slides on what’s expected of you in the labs.
Important preliminary remarks
We use jupyter notebook or jupyter lab, potentially with Visual Studio Code, for all practical
sessions. Some important points to remember :
- Loading -
import sklearn # import a package
import numpy as np # import a package under an alias
import matplotlib.pyplot as plt # import a submodule with an alias
from sklearn import linear_model # import a submodule
from os import mkdir # import a peculiar function
- Using standard help -
When facing a difficulty, you are strongly encouraged to refer to the online documentation of pandas,
numpy, etc. It should become a reflex to look for the answer in the doc or on stackoverflow.
linear_model.LinearRegression? # to get some help on the LinearRegression object
- Package versions -
print(np.__version__) # to get a package version
Strings
1) From a string containing all the alphabet letters, generate the string cfilorux using slicing (notice
the pattern : this is the 3rd letter, then the 6th, then the 9th, etc). Do the same for the strings
vxz and zxvt (again, notice the patterns). Note : don’t type the whole alphabet yourself, use the
string module.
2) Declare a string variable " XHEC DataScience for Business ". Make it all lowercase. Remove
spaces at the beginning and at the end, but not between words. Replace all e’s with E’s.
page 1
3) Display the number π with 9 decimal digits (np.pi). Don’t cast a number to string, and don’t use
round : use Python’s string formatting instead (either the format method, either the % operator,
either an f-string – the latter is considered more modern).
4) Count the number of occurrences of each character in the string s = "HelLo WorLd!!" (in real
life, you should use a collections.Counter ; here, you are asked to code the method yourself).
Output a dictionary that to each character associates the number of occurrences in this string. In
this question, we consider that lower and upper case characters are the same (e.g. your dictionary
should not have both a L and a l entry).
Fast computations with numpy ; basic plots.
Hint : useful functions : np.arange, np.allclose, np.all, np.linalg.norm for instance. The
whole numpy.linalg module contains interesting Linear Algebra functions. numpy.random contains func-
tions to generate arrays of (pseudo-) random numbers.
In all this section, unless asked explicitly, you cannot use for/while loops (they are slow in native
python).
5) Compute 0.1 + 100 - 100. Using ==, check if it is equal to 0.1. Comment. Compare the two
floating point numbers again with an appropriate numpy function.
6) Create a list (resp. a numpy array) containing all square numbers from 1, 4, ... to 121, using a for
loop (resp. only numpy). Why should you use arrays instead of loops whenever possible ?
7) Create an array containing integers from 2 to 14 by step of 3 (2, 5, 8, ...). Create an array with 15
equispaced valued from 0 to 1 included. Use numpy built-in functions.
ś8 2
8) Compute 2 k“1 4k4k2 ´1 (approximate 8 by a large number n) using a for loop. Propose a ver-
sion without loop, using only numpy (see numpy.prod). Measure the time taken by both versions
using time.time(). You should display the results with a relevant number of significant digits, e.g.
not 0.002487976589749873 seconds. Use the ipython magic %timeit again to measure time of one
version. Why is it better than time.time ?
9) (row and column vectors, aka numpy only knows 1D arrays) Compute the dot product (aka scalar
product) of np.arange(5) and np.ones(5). What is the shape of np.arange(5) ? How many di-
mensions does this array have ? What is the shape of its transpose (use .T) ? What does transposing
1D arrays do ?
10) What does reshape do in M = np.arange(12).reshape(2, 6) ? What does M[:, ::3] do ? What
happens when you do np.arange(3) * np.arange(4)[:, np.newaxis] (this powerful tool is
known as broadcasting)
11) Create a random matrix M P R5ˆ6 with coefficients taken uniformly (and independently) in r´1, 1s.
Substract to each even column of M (say M[:, 0] is even), twice the value of the following (uneven)
column.
12) Replace the negative values in M by 0. Compute the mean of each row of M . Substract to each
row of M , its mean.
13) Create a random matrix M P R5ˆ10 with coefficients taken uniformly (and independently) in r´1, 1s.
Test whether G “ M J M is symmetric semi-definite positive, and that its eigenvalues are strictly
positive. Compute the rank of G. Compute the Euclidean norm of G. Compute the operator norm
of G (aka spectral norm, aka Schatten 2-norm). Compute the standard deviation of each column of
G.
14) Plot the functions x ÞÑ xd on the interval r´1, 2s for d P t2, 3, 4u with a decent resolution. Put a
xlabel, a ylabel, legend the 3 curves with d “ 2, d “ 3, d “ 4 respectively. Put a title.
Numpy advanced behavior
Numpy broadcasting
In this exercise, you can use neither lists nor for loops. You should use only numpy’s operations,
which are fast. An introduction to broadcasting is available here : https://numpy.org/doc/stable/
user/basics.broadcasting.html.
page 2
15) Create an array with integers 1, 3, ..., 19. Subtract its mean to it. Observe than you can thus
subtract a number to an array, even though they do not have the same shape. Create an array with
3 lines and 4 columns, such that arr[i, j] = 4 * i + j (it thus contains integers from 0 to 11).
reshape will help.
16) Now, we subtract vectors to 2D arrays, using broadcasting. Take the previous (3, 4) array, and
subtract its column wise mean to it (easy). Subtract its row wise mean to it (technically more
challenging the first time, you can add an axis to a 1D array with arr[:, None]).
17) Using broadcasting and np.arange, create an array of shape (3, 5) such that arr[i, j] = i * j.
Value and reference types
18) Create a variable a equal to 1000. Check the address in memory of the variable with the builtin id
function. Create a second variable b equal to 1000. What is the address in memory of b ? Create a
third variable c equal to a, check its address in memory. Do a += 1. How does it affect the values
of the three variables ? Why ?
19) Do the same but this time using a = np.array([0, 1]), b = np.array([0, 1]), c = a. What is
going on ?
20) (passing by value/passing by reference) Define a function f as follows : def f(a): a += 1. Call it
on a = 1, then on a = np.ones(10). For both cases, check the value of a after calling f on it.
What’s the reason for this behavior ?
21) For two arrays a = np.zeros(10), b = np.ones(10), what’s the difference between doing a = b
and a[:] = b ? What’s the difference between a = a + 1 and a += 1 ?
page 3
