This project is aimed at learning the different procedures when looking for a target state. In this project we have covered three different search methods:
- Breadth-first search (BFS)
- Depth-first search (DFS)
- Breadth-sorted search (BSS) [Actually it is more like Branch and Bound]
- Combining sorted search with heuristic
This methods builds a tree by examining the nodes of the graph. The process is referred as spanning tree. (Refer to https://en.wikipedia.org/wiki/Spanning_tree for further details)
The last search method is key to retrieve the shortest route. It 'keeps track' of every path cost and sort them based upon this criteria, thus determining efficient routes.
Such searches have been implemented using two lists:
- Open List (OL): stores nodes that are pending to be analysed. Depending on the search method a new set of nodes are inserted by its front or back. There is a subtle modification regarding BSS.
- Close List (CL): a list comprised of nodes already analysed. If a node that is being analysed is contained within this list, this node is discarded. The node is ditched because a loop has been discovered.
The code must retrieve the shortest route given two geographic points, based upon this map.
This section shows what the output should be chosen two geographical points and a search method:
From Zerind to Sibiu (See the map)
First letter of each city is used to represent the city:
Zerind = Z; Sibui = S; Rimnicu Vilcea = R
- Using BFS: all nodes belonging to a certain depth are first scanned before diving into the next level
OL= {Z}
CL = {}
Pop Z, is Z the final state (ie, S)? No. Is Z in CL? No
OL = {A,O}
CL = {Z}
Pop A, is A S? No. Is A in CL? No
OL = {O, Z, S}
CL = {Z, A}
Pop O, is O S? No. Is O in CL? No
OL = {Z, S, Z, S}
CL = {Z, A, O}
Pop Z, is Z S? No. Is Z in CL? Yes
OL = {S, Z, S}
Pop S, is S S? Yes.
Output: {S, A, Z}
Every node contains a reference to its parent.
A visual scheme is:
(This awesome depiction was made using https://sketch.io/sketchpad/)
- Using DFS: a node is scanned until the lowest leaves (a node with no children)
OL = {Z}
CL = {}
Pop Z, is Z the final state (ie, S)? No. Is Z in CL? No
OL = {O,A}
CL = {Z}
Pop O, is O S? No. Is O in CL? No.
OL = {O,S,A}
CL = {Z, O}
Pop O, is O S? No. Is O in CL? Yes.
Ol = {S,A}
Pop S, is S S? Yes.
Output: {S, O, Z}
- Using BSS: in every insertion nodes within OL are sorted based upon path_cost attribute (this sttr. is accumulative)
Chosen points: from Zerind to Sibiu (Z, S)
Since in every expansion all nodes are sorted, the insertion does not matter at all.
OL = {Z}
CL = {}
Is OL sorted? Pop Z, is Z S? No. Is Z in CL? No.
OL = {O,A}
CL = {Z}
Is OL sorted? yes.
Z-O = 71
Z-A = 75
Pop O, is O S? No. Is O in CL? No.
OL = {A,Z,S}
CL = {Z,O}
Is OL sorted? yes.
Z-A=75
Z-O-Z = 71+71 = 142
Z-O-S = 71+151 = 222
Pop A, is A S? No. Is A in CL? No.
OL = {Z,S,Z,S}
CL = {Z,O,A}
Is OL sorted? No.
Z-O-Z = 142
Z-O-S = 222
Z-A-Z = 75+75 = 150
Z-A-S = 75+140 = 215
sort, OL = {Z, Z, S, S}
Pop Z, Is Z S? No. Is Z in CL? Yes.
OL = {Z,S,S}
Pop Z, Is Z S? No. Is Z in CL? Yes.
OL = {S,S}
Pop S, Is S S? Yes.
Output: Z-A-S (path cost = 75+140 = 215)
As you might have noticed, they are greedy algorithms, hence not suitable for large maps.
- Using heuristic
This method sorts not only based on path_cost but the heuristic of every node (city). You can understand the heuristic
as the distance between such city and the final state (distances are measured roughly). There is a method
within GPSProblem called h(self, node) which retrieves such distance between node and the final state in question.
Obviously the expanded nodes are less using the heuristic method rather than the BSS method.