首页 > > 详细

代做EE 496代写Python实验

Evolutionary 
Art 
METU EE 496 
Introduction to 
Computational Intelligence 
Homework 2 - Evolutionary Algorithms 
Due: 23:55, 19/04/2020 
Late submissions are welcome, but penalized according to the following policy: 
• 1 day late submission: HW will be evaluated out of 70. 
• 2 days late submission: HW will be evaluated out of 50. 
• 3 days late submission: HW will be evaluated out of 30. 
• 4 or more days late submission: HW will not be evaluated. 
You should prepare your homework by yourself alone and you should not share it with other students, 
otherwise you will be penalized. 
Homework Task and Deliverables 
In this homework, you will perform experiments on evolutionary algorithm and draw conclusions from 
the experimental results. The task is to create an image made of filled circles, visually similar to a given 
RGB source image (painting.png). 
The implementations will be in Python language and you will be using OpenCV package to draw the 
images. You can install it using pip install opencv-python command. You can use import cv2 to 
use it in your code. 
You should submit a single report in which your answers to the questions, the required experimental 
results (plots, visualizations etc.) and your deductions are presented for each part of the homework. 
Moreover, you should append your Python codes to the end of the report for each part to generate 
the results and the visualizations of the experiments. Namely, all the required tasks for a part can be 
performed by running the related code file. The codes should be well structured and well commented. 
The submissions lacking comments will be simply not evaluated. 
The report should be in portable document format (pdf) and named as hw2 name surname eXXXXXX 
where name, surname and X s are to be replaced by your name, surname and digits of your user ID, 
respectively. 
1 The Evolutionary Algorithm 
The pseudocode for the algorithm is as follows: 
Initialize population with individuals each having genes 
While not all generations () are computed: 
Evaluate all the individuals 
Select individuals 
Do crossover on some individuals 
Mutate some individuals 
1.1 Individuals 
Each individual has one chromosome. Each gene in a chromosome represents one circle to be drawn. 
Each gene has at least 7 values: 
• The center coordinates, (x, y) 
• The radius, radius ∈ Z+ 
• The color (red, R ∈ Z, green, G ∈ Z, blue, B ∈ Z, alpha, A ∈ R) where (R,G,B) ∈ [0, 255]3 and 
A ∈ [0, 1] 
The order of the tasks to perform circle drawing is important. The circles should be drawn in the 
descending order of their radii. The first circle to be drawn is the one with the largest radius and the 
last circle to be drawn is the one with the smallest radius. The genes should be sorted accordingly. 
The center does not have to be within image boundaries. However, if a circle is not within the image (it 
lies outside completely), the corresponding gene should be reinitialized randomly until it is. 
1.2 Evaluation 
In order to evaluate an individual, its corresponding image should be drawn first. Note that the chro- 
mosome order is important. The pseudocode is as follows: 
Initialize  completely white with the same shape as the . 
For each gene in the chromosome: 
overlay <- image 
Draw the circle on overlay. 
img <- overlay x alpha + image x (1 { alpha) 
The fitness function, F , is 
F = − 
∑ 
k∈{R,G,B} 
∑ 
06i06j(source imagei,j,k − imagei,j,k)2 (1) 
where source imagei,j,k is the pixel value, at i 
th column and jth row, in channel k in the source image. 
Similarly, imagei,j,k represents a pixel value in the individual’s image. 
1.3 Selection 
number of best individuals will advance to the next generation directly. The selection of 
other individuals is done with tournament selection with size . 
1.4 Crossover 
number of individuals will be used for crossover. The parents are chosen among the best 
individuals which do not advance to the next generation directly. Two parents will create two children. 
Exchange of each gene is calculated individually with equal probability. The probabilities of child 1 
having genei of parent 1 or parent 2 have equal probability, that is 0.5; child 2 gets the genei from the 
other parent which is not chosen for child 1, where 0 6 i < . 
Table 1: Parameter configurations to be experimented. 
5 10 20 50 75 
10 25 50 100 150 
2 5 10 20 
0.05 0.2 0.4 
0.2 0.4 0.6 0.8 
0.1 0.2 0.5 0.8 
guided unguided 
1.5 Mutation 
The mutation is governed by . While the generated random number is smaller than 
a random gene is selected to be mutated (same as in N Queen Problem in the lecture 
notes). All individuals except the elites are subject to mutation. 
There are two ways a gene can be mutated: 
• Unguided 
– Choose completely random values for x, y, radius,R,G,B,A. 
• Guided 
– Deviate the x, y, radius,R,G,B,A around their previous values. 
∗ x–width4 < x′ < x + width4 
∗ y– height4 < y′ < y + height4 
∗ radius− 10 < radius′ < radius + 10 
∗ R− 64 < R′ < R + 64 
∗ G− 64 < G′ < G + 64 
∗ B − 64 < B′ < B + 64 
∗ A− 0.25 < A′ < A + 0.25 
– The values should be corrected to a valid one in case they are not. 
2 Experimental Work 
You will experiment on several different hyperparameters. Namely, 
• Number of individuals () 
• Number of genes () 
• Tournament size () 
• Number of individuals advancing without change () 
• Number of parents to be used in crossover () 
• Mutation probability () 
• Mutation type (guided or unguided) 
To see the effect of each hyperparameter, run the algorithm for each value in a row given in Table 
1 while using default values for the other hyperparemeters. The default values are bolded. Use < 
num generations >= 10000. 
For each hyperparameter, explain its effect and give the value which produces the best result. For each 
experiment, you need to include: 
• fitness plot from generation 1 to generation 10000 
• fitness plot from generation 1000 to generation 10000 
• the corresponding image of the best individual in the population at every 1000th generation. 
3 Discussion 
Suggest three changes to this evolutionary algorithm which may provide faster and/or better convergence. 
The changes should not be only using a different value for a hyperparameter. Show the result empirically 
and explain. 
4 Remarks 
You may find the following useful... 
1. Because of the nature of assignment in Python, be careful how you use them. Consider the following: 
a = [[1, 2], [3, 4]] 
b = [a[0], a[0]] 
b[0].append(5) 
print(b) # prints [[1, 2, 5], [1, 2, 5]] 
print(a) # prints [[1, 2, 5], [3, 4]] 
At times, you may want to use deepcopy from copy module. 
2. You can use cv2.circle to draw circles and cv2.addWeighted to blend images. 
3. Because the underlying type for the image is np.uint8, be careful of possible overflows. 
4. You are advised to use classes to implement genes, individuals and populations. It will make 
debugging easier. 
5. You are strongly advised not to do your homework on the last day of submission. The experiments 
may take a while to finish. 

联系我们
  • QQ:99515681
  • 邮箱:99515681@qq.com
  • 工作时间:8:00-23:00
  • 微信:codinghelp
热点文章
程序代写更多图片

联系我们 - QQ: 99515681 微信:codinghelp
© 2014 www.7daixie.com
程序代写网!