INSTITUT FÜR NEUROINFORMATIK

Artificial Neural Networks

Introduction to python
Richard Görler
WHAT'S IT ALL ABOUT?
These slides are meant as a reference. We will not go through all of them in a
presentation.

Tutorial contents:

1. System setup

2. How to python
- Operators, data types, built-in functions
- Functions, modules and classes
- Control structures (conditionals and loops)
- Pythonians: args/kwargs, list comprehensions and iterators

3. Vector / matrix calculations with numpy

4. Plotting with pyplot

INSTITUT FÜR
2 CONTENTS NEUROINFORMATIK
SYSTEM SETUP
Version information
We will be using Python 3.6.8 with the following packages:

numpy 1.15.4
scipy 0.19.1
matplotlib 2.0.2
sklearn (scikit-learn) 0.19.1
tensorflow 1.12.0
keras 2.2.4

Code will most probably work on other versions as well, but we cannot guarantee

You can use the virtual environment from Moodle. Instructions are provided
below.

INSTITUT FÜR
4 How to python NEUROINFORMATIK
Using the Anaconda environment

●
Download Miniconda (https://docs.conda.io/en/latest/miniconda.html, Python 3.* version)
or Anaconda (https://www.anaconda.com/distribution/, Python 3.* version). In order to
save time and disk space, Miniconda is recommended.
●
Install Anaconda.
Windows: You do not need to add python to your system path in the process.
●
Download the file annvenv.yml from Moodle
●
Open a terminal (Windows: the Anaconda Prompt)
●
In the terminal / Anaconda Prompt, navigate to the directory in which annvenv.yml is
located (use cd dir/to/file)
●
Create the environment with
conda env create -f annvenv.yml
●
Activate it with
conda activate annvenv

Next time you only need to do the last step (activation)

INSTITUT FÜR
5 How to python NEUROINFORMATIK
Alternative

Of course you can install python and the packages without using the Anaconda
environment.

For instance on Linux systems:

●
Use your package manager to install python
(e.g. on Ubuntu apt-get install python3.6)
●
Use python's package manager pip to install the python packages
(pip install [package]==[version])

INSTITUT FÜR
6 How to python NEUROINFORMATIK
Multiple python versions

If you already have a different version of python installed, just installing

python3.6 might not work, or make package management difficult.
Consider the following options

●
Use your installed versions – as long as they are newer than the ones we use
in this course it will work most probably (and as long as the primary version
number is identical)
●
Use pyenv to manage multiple python versions. It works on linux and mac.
Information on installation can be found here:
https://github.com/pyenv/pyenv-installer,
for usage see here https://github.com/pyenv/pyenv#table-of-contents and here
https://github.com/pyenv/pyenv/blob/master/COMMANDS.md

INSTITUT FÜR
7 How to python NEUROINFORMATIK
IDE: Spyder

Spyder is recommended as an IDE.

Spyder 3.3.6 is included in the Anaconda environment.

INSTITUT FÜR
8 How to python NEUROINFORMATIK
CIP rooms in ID

On the computers in CIP1 and CIP2 Anaconda and the required packages are
already installed. You do not need to create or activate the environment.

INSTITUT FÜR
9 How to python NEUROINFORMATIK
HOW TO PYTHON

Operators, data types, built-in functions

Operators

Arithmetic

+ - * / // ** %

Relational

== != < <= > >= is

Assignment

= += -= /= […]

Logical

and or not

INSTITUT FÜR
11 How to python NEUROINFORMATIK
Data types
are implicit in python

Numbers

int: 42 float: 42.0 (can be inf or nan)

Lists and strings

[1,2,3] 'Hello' "Hello"

String and list concatenation

'Hello ' + 'world' [1,2,3] + [4,5,6]

Others

bool: True / False NoneType: None Inf, NaN

INSTITUT FÜR
12 How to python NEUROINFORMATIK
Accessing list elements
lst = [1, 2, 3, 4, 'potato']

First element / last element

lst[0] lst[-1]

List slicing

lst[start, stop, step]

start: first element (index) to include
stop: first element (index) to exclude
step: step size

Examples
lst[4:1:-2]
lst[2:]
lst[:-1]
lst[::-1]

INSTITUT FÜR
13 How to python NEUROINFORMATIK
More on lists

lst1 = [1, 2, 3]

Test if item is contained

3 in lst1
4 in lst1

In-place manipulation: extend / append

lst1.append([4, 'potato'])
lst1.extend([4, 'potato'])

Other useful functions

lst1.pop() or lst1.pop(2)
lst1.index('potato')
etc.

INSTITUT FÜR
14 How to python NEUROINFORMATIK
Dictionaries

dct = {'one': 1, 'two': 2} or dct = dict( two=2)

Dictionary access

dct['one']

dct.keys() dct.values()

Note:
You can also use numbers or tuples, e.g. (1, 'a') as keys

INSTITUT FÜR
15 How to python NEUROINFORMATIK
Built-in functions
Console output

print()

Type conversion

str() int() float()

Numerical

abs()

List properties

len() min() max() sum()

Other

range(start, stop, step)

INSTITUT FÜR
16 How to python NEUROINFORMATIK
HOW TO PYTHON

Functions, modules and classes

Scripts

File hello.py:

print("Hello World")

Navigate to the directory containing your script

python hello.py

INSTITUT FÜR
18 How to python NEUROINFORMATIK
Defining functions

File fun.py:

def inc(x=0): (var=default value)

y = x + 1
return y

Indentation is used to define hierarchical blocks. No start / end markers.

Best practice is to use one tab.

Start python from the directory containing your file

import fun
fun.inc(1337)
fun.inc()

INSTITUT FÜR
19 How to python NEUROINFORMATIK
Modules
Directory structure:
fun.py
mod/__init__.py (double underscore)
mod/fun2.py

Script fun2.py:

def dec(x):
y = x - 1
return y

Now you can do:

import fun
from mod import fun2
a = fun.inc(1337)
fun2.dec(a)

INSTITUT FÜR
20 How to python NEUROINFORMATIK
Classes
File potato.py

class Potato(): (inheritance)

def __init__(self, weight=100):

self.weight = weight

def report(self):
print("This potato weighs " + str(self.weight) + " grams.")

Now you can do:

import potato
p1 = potato.Potato()
p2 = potato.Potato(250)
p1.report()
p2.report()

INSTITUT FÜR
21 How to python NEUROINFORMATIK
HOW TO PYTHON

Control structures
if statements

def sign(x):
if x < 0:
return -1
elif x > 0:
return 1
else:
return 0

INSTITUT FÜR
23 How to python NEUROINFORMATIK
for loops

for w in ['cat', 'window', 'defenestrate']:

print(w)

or

for i in range(5):
print(i)

enumerate can be useful

for i, w in enumerate(['one', 'two', 'three']):

print(i, w)

INSTITUT FÜR
24 How to python NEUROINFORMATIK
while loops

i = 0
while i < 10:
print(i)
i += 1

Note:
For both types of loops, the break and continue statements are available.

INSTITUT FÜR
25 How to python NEUROINFORMATIK
HOW TO PYTHON

Pythonians
*args, **kwargs

You can use pointers to lists or dictionaries to set function arguments

def add(x, y, z=1):

return x + y + z

lst = [3, 4, 5]
dct = dict(x=3, a=10, y=4, z=5)
add(*lst)
add(**dct)

Lists provide arguments without keywords (*args)

while dicts provide arguments with keywords (**kwargs)

INSTITUT FÜR
27 How to python NEUROINFORMATIK
List comprehensions

Make a new list from a list

lst = list(range(10))
[abs(n-5) for n in lst]

We can also use conditionals

[abs(n-5) for n in lst if n != 5]

INSTITUT FÜR
28 How to python NEUROINFORMATIK
Iterators
Make an iterator from a list

it = iter([1, 8, 42, 1337, "over 9000"])

next(it)
next(it)
…

Generators work similarly, but programatically:

def gen():
i = 0
while True:
yield i
i += 1

g = gen()
next(g)
next(g)
...

INSTITUT FÜR
29 How to python NEUROINFORMATIK
MATRIX CALCULATIONS
WITH NUMPY
What is numpy?
Package containing functions for vector calculations.

import numpy as np

Basic data structure: array

arr = np.array([2, 3, 5, 7, 11, 13, 17, 19])

arr
type(arr)

Access works the same as for lists

arr[0]

Number type and conversion

arr.dtype
arrF = arr.astype(float)

INSTITUT FÜR
31 Matrix calculations with numpy NEUROINFORMATIK
Defining arrays
arr1 = np.arange(5)
arr2 = np.zeros(5)
arr3 = np.ones(5)

Arrays can have more than one dimension

arr11 = np.array([[1,3,5,7,9], [2,4,6,8,10]])

arr22 = np.zeros((2,5))
arr33 = np.ones((2,5))
arrEy = np.eye(5) unit matrix

Shape

arr10 = np.arange(10)
arr10.shape
arr10.reshape((2, 5))

INSTITUT FÜR
32 Matrix calculations with numpy NEUROINFORMATIK
Random numbers

2-by-5 matrix containing random floats between 0 and 1:

rand1 = np.random.rand(2,5)

10-element vector containing floats drawn from a "standard normal" distribution:

rand2 = np.random.randn(10)

Many more options, see the documentation!

Google: "numpy random"

INSTITUT FÜR
33 Matrix calculations with numpy NEUROINFORMATIK
Array indexing
Give indices for both dimensions

arr11[0, 4]

Access an entire column using a colon

arr11[:, 2]

Fancy indexing

Index an array with an array.

arr = np.array([2, 3, 5, 7, 11, 13, 17, 19])

ind = np.array([0,2,4,6])
arr[ind]

You can also edit elements accessed that way

arr[ind] = 0

INSTITUT FÜR
34 Matrix calculations with numpy NEUROINFORMATIK
Operations on arrays

Operators are element-wise:

a = np.array([[1,3,5], [-5,-7,-9]])
b = np.array([[2,4,6], [6,8,10]])
a+b
a*b
a<b
b-5
etc.

numpy element-wise functions (examples)

np.abs(a)
np.exp(b)

INSTITUT FÜR
35 Matrix calculations with numpy NEUROINFORMATIK
Basic data processing
a = np.array([[1,3,5], [-5,-7,-9]])
np.max(a)
np.min(a)
np.mean(a)
np.std(a) Note that this is population std. For sample std use np.std(a, dof=1)

Select axis along which to process

np.mean(a, axis=0)
np.mean(a, axis=1)

Load and save arrays

np.save("a.npy", a)
a1 = np.load("a.npy")

INSTITUT FÜR
36 Matrix calculations with numpy NEUROINFORMATIK
np.nonzero()

Get indices of non-zero elements

arr0 = np.array([[0,1,2,0,3,0],[0,4,5,0,0,6]])
nnz = np.nonzero(arr0)
nnz
arr0[nnz]

Use it to find where condition is met

arr10 = np.arange(10)
nnz10 = np.nonzero(np.abs(arr10-5) > 2)
nnz10
arr10[nnz10]

INSTITUT FÜR
37 Matrix calculations with numpy NEUROINFORMATIK
Also useful

Number of elements that meet condition

np.sum(np.abs(arr10-5) > 2)

Compare entire arrays

np.array_equal(a, b)

Check for invalid data

t1 = np.isnan(a)
t2 = np.isinf(b)

any / all

np.any(t1)
np.all(t2)

INSTITUT FÜR
38 Matrix calculations with numpy NEUROINFORMATIK
Inner and outer product

(inner
product)

(outer
product)

np.inner() / np.dot() and np.outer()

For matrix multiplication, use np.matmul(A,B) or A @ B

INSTITUT FÜR
39 Matrix calculations with numpy NEUROINFORMATIK
PLOTTING WITH PYPLOT
Plot
import matplotlib.pyplot as plt
x = np.arange(20)-10
y = np.abs(x)
plt.plot(x,y)
plt.show()

Subplot

plt.subplot(1,2,1)
plt.plot(x,y)
y2 = np.exp(x)
plt.subplot(1,2,2)
plt.plot(x,y2)
plt.show()

INSTITUT FÜR
41 Plotting with pyplot NEUROINFORMATIK
Scatter (and more...)

xr = np.random.randn(1000)
yr = np.random.rand(1000)
plt.scatter(xr,yr,label="Data")

plt.plot([-3.5, 3.5], [0.5, 0.5],

color='g', label="A line")

plt.xlabel("x-axis")
plt.ylabel("y-axis")
plt.title("This is very useful")

plt.legend()

plt.savefig("example.pdf")
plt.show()

INSTITUT FÜR
42 Plotting with pyplot NEUROINFORMATIK
Bar

x = [0, 1]
y = [0.5, 0.7]
std = [0.2, 0.25]
plt.bar(x, y, width=0.8, yerr=std)
plt.show()

INSTITUT FÜR
43 Plotting with pyplot NEUROINFORMATIK
Imshow

m = np.eye(10)
plt.imshow(m, interpolation='none', cmap="Greys")
plt.show()

INSTITUT FÜR
44 Plotting with pyplot NEUROINFORMATIK
More options

plt.imshow(m, interpolation='none',
cmap="Greys")
plt.xlim([-10,20])
plt.ylim([15,-5])
plt.xticks([0,4,8,12])
plt.yticks([1,5])
plt.show()

plt.axis('off')

INSTITUT FÜR
45 Plotting with pyplot NEUROINFORMATIK
DONE!