sunxy88 / xispyrat

Homework for Algorithm and discrete mathematics

Created by Xi SONG in IMT Atlantique, Brest, 05/09/2019

---

This repository is created for the course Algorithm and Discrete Mathematics in FISE A1S1.

All code for the creation of my AI will be showed in thie repository.

Talk is cheap! I will show you my code.

• Lab1: Program a rat that moves randomly while avoiding walls

  • Code admitted: improved_random.py at 06/09/2019, DataAnalyse.py at 11/09/2019
  • Question found:
    • In the project PyRat, a file named random.py which may cause problems in importing library numpy
  • Skill learned:
    • Meet library numpy
    • Meet library matpltlib
  • Lessons:
    • Read source code pyrat.py is necessary

• Lab2: Implement a BFS under the condition of unweighted graphs

  • Code admitted: BFS_move.py at 15/09/2019, DFS_move.py at 16/09/2019, IDDFS_move.py at 21/09/2019
  • With the will to automatically analyse the code, update AutomaticGetStatistics.sh (ATTENTION: This script may not be useful in Windows)
  • Question found:
    • (Solved) Every turn, main program calls function turn, however I wrote the code with the imagination that only one call of this function and the pause of this thread. So it is necessary to make this code correct. -> Use a file instructions.txt to guide the rat
    • (Solved) While coding the DFS, the recursion can not be terminated. -> Have to clearly understand the termination condition for a recursion function
    • (Solved) In beam search, which cost fonction should be chosen for the heuristic cost? Maybe Euclidean distance is a good choice
    • (Solved) Also in beam search, how should we choose an approprate beam width, for 3 at most? For this reason, I am wandering if it is practical to use beam search in this game. -> It is not interesting
    • (Solved) In shell script, how to automatically kill the process elegantly? -> No need, pyrat.py offer a setting to do the test
  • Lessons learned:
    • Before starting to code, it is vital to understand clearly the demand
    • In every recursion function, we must shrewdly consider all the possible conditions for the determination of a recursion or we will get lost and a stack overflow
    • Use Shell script to automatically do something interesting

• Lab3: Implement Dijkstra's algorithm

  • Code admitted: Dijkstra_move at 25/09/2019, AutomaticGetStatistics.sh at 26/09/2019
  • Question found:
    • (Solved) Whilte trying to use heapq.heappush(priorityQ, (, )), the order of the element containing tuple is not logical. -> Actually, heapq.heappush() use __ls__() to decide the order in a list
  • Lessons learned:
    • Meet heapq, and understand how it decides the order of a heap
    • Reload operators
    • In binary balance tree (AVL Tree), two kinds of rorations (hard to under stand and click here to know more) are used to keep the balance factor under 2

• Lab4: Implement bruteforce and backtracking algorithm to get multiple pieces of cheese

  • Code admitted: BruteForce_move.py at 01/10/2019, Backtracking_move.py at 01/10/2019 and NaiveGreedy.py

• Lab5: Implement heuristic to speed up the preprocessing

  • Code admitted: GeneticAlgorithm.py at 10/10/2019
  • Question found:
    • Results of GeneticAlgorithm is not that good, have to tune parameters to make it faster and solutions better
  • Lessons learned:
    • Genetic algorithm can provide a solution of a question even we only know how to judge a solution

• Lab6: Refactor the genetic algorithm in speeding up to get a metagraph, so it is practical to use genetic graph in maze where there are 41 cheeses.

  • Code admitted: Winner.py at 01/11/2019
  • Question found:
    • The implementation of dijkstra before (in which I use heapq) is in a way much slower than the implementation now, maybe there is something redundant calculations.
  • Lessons learned:
    • It is meaningless to optimize your code before it is done, and also not useful to optimize some minor details. Doing so only makes your code hard to read and confusing.
  • Work could be done:
    • To tune the parameters in the genetic algorithm could be annoying, should find a way to do it elegently and efficiently.

---

Xidian University / IMT Atlantique