I'm trying to implement a Neighbor Joining algorithm, at the moment I have got it working as it should, calculating the correct lengths at each step and outputting the correct values.
However, I am struggling with obtaining the final output of the algorithm, I need it to output the overall calculated matrix representation because I would like to represent it visually as a graph. Through each iteration of the main loop of the algorithm I get a sub group of nodes that goes back into the start of the algorithm, but I don't believe that this subgroup could be used because it contains redundant information that I can't really specify whether or not will be needed in the final representation.
I am using this algorithm here: http://en.wikipedia.org/wiki/Neighbor_joining#The_algorithm
Any help would be fantastic and I can provide more information if needed, thanks.
I have read the link you provided and it seems to me that you do need the information.
Each step of the algorithm merges 2 nodes into 1, making your distance matrix smaller until everything is merged. You need to remember the distances of the nodes you merge to their resulting node. If you merge A and B, then the columns/rows of your distance matrix are replaced by a column/row belonging to a new node, u. You need to remember the distances of A and B to u.
After everything is merged, you should have all distances of all nodes that have to be connected and you can start visualizing.
Related
I was trying to implement Karger's min cut algorithm in the same way it is explained here but I don't like the fact that at each step of the while loop we can pick an edge with it's two endpoints already in a supernode. More specifically, this part
// If two corners belong to same subset,
// then no point considering this edge
if (subset1 == subset2)
continue;
Is there a quick fix for avoiding this problem?
It might help to back up and think about why there’s a union-find structure here at all and why it’s worth improving on the continue statement.
Conceptually, each contraction performed changes the graph in the following way:
The nodes contracted get replaced with a single node.
The edges incident to either node get replaced with an edge to the new joint node.
The edges running between the two earlier nodes get removed.
The question, then, is how to actually do this to the graph. The code you’ve found does this lazily. It doesn’t actually change the graph when the contraction is done. Instead, it uses the union-find structure to show which nodes are now equivalent to one another. When it samples a random edge, it then has to check whether that edge is one of the ones that would have been deleted in step (3). If so, it skips it and moves on. This has the effect that early contractions are really fast (the likelihood of picking two edges that are part of contracted nodes is very low when few edges are contracted), but later contractions might be a lot slower (once edges have started being contracted, lots of edges may have been deleted).
Here’s a simple modification you can use to speed this step up. Whenever you pick an edge to contract and find that its endpoints are already connected, discard that edge, and remove it from the list of edges so that it never gets picked again. You can do this by swapping that edge to the end of the list of edges, then removing the last element of the list. This has the effect that every edge processed will never be seen again, so across all iterations of the algorithm every edge will be processed at most once. That gives a runtime of one randomized contraction phase as O(m + nα(n)), where m is the number of edges and n is the number of nodes. The factor of α(n) comes from the use of the union-find structure.
If you truly want to remove all semblance of that continue statement, an alternative approach would be to directly simulate the contraction. After each contraction, iterate over all m edges and adjust each one by seeing whether it needs to remain unchanged, point to the new contracted node, or be removed altogether. This will take time O(m) per contraction for a net cost of O(mn) for the overall min cut calculation.
There are ways to speed things up beyond this. Karger’s original paper suggests generating a random permutation of the edges and using binary search over that array with a clever use of BFS or DFS to find the cut produced in time O(m), which is slightly faster than the O(m + nα(n)) approach for large graphs. The basic idea is the following:
Probe the middle element of the list of edges.
Run a BFS on the graph formed by only using those edges and see if there are exactly two connected components.
If so, great! Those two CCs are the ones you want.
If there is only one CC, discard the back half of the array of edges and try again.
If there is more than one CC, contract each CC into a single node and update a global table indicating which CC each node belongs to. Then discard the first half of the array and try again.
The cost of each BFS is O(m), where m is the number of edges in the graph, and this gives the recurrence T(m) = T(m/2) + O(m) because at each stage we’re throwing away half of the edges. That solves to O(m) total time, though as you can see, it’s a much trickier way of coding this algorithm up!
To summarize:
With a very small modification to the provided code, you can keep the continue statement in but still have a very fast implementation of the randomized contraction algorithm.
To eliminate that continue without sacrificing the runtime of the algorithm, you need to do some major surgery and change approaches to something only marginally asymptotically faster than keeping the continue in.
Hope this helps!
Isn't it always better when searching for shortest path to use for connected nodes lists instead of grid?
When using grid, you have to iterate over the grid every time, whereas using lists saves lots of time.
With adjacency matrix usually each check costs you O(n) time. It may be a bit slower than a list of connected nodes. However, you can do some fancy stuff with it. For example, if you want to delete a lot of edges, you can do it in O(1) using adjacency matrix (it may take a lot longer using a list of nodes depending on what data structure you use for it). Adjacency matrix is also a matrix. What do I mean by that? If you want to check in how many ways you can get from node A to node B in k steps, you can raise this matrix to the power of k, which is impossible to do with a list.
How to partition a graph into possibly overlapping parts such that any vertex is contained in a part at which it has at least distance k from the Boundary?
The problem arises in cases where the whole graph can not be loaded into a single machine
because there is not sufficient memory. So another requirement is that the partition has
somehow an equal number of vertices.
Are there any algorithms that try to minimize the common vertices between parts?
The use case here is this: You want to perform a query starting from an initial vertex that you know will require maximum k traversals. Having a part that contains all the vertices of this query results in zero
network utilization.
The problem thus is to reduce the memory overhead of such a partition.
Any books I should read?
I found this which looks promising:
http://grafia.cs.ucsb.edu/sedge/docs/sedge-sigmod12-slides.pdf
final edit: It is no coincidence that google decided to use a Hash partition.
Finding a good partition is difficult. I ll go with a hash partition as well and hope
that the data center has good network bandwidth.
You can use a breadth first search to get all the nodes that are only k distance away from the node in question, starting with the node itself. When you reach k away from the origin, you can end the search.
Edit:
Use a depth first search to assign a minimum distance from boundary property to each node. Once you have completed the depth first search, a simple iteration through the nodes should provide the partitions. For example, you can create a table that stores the minimum distance from boundary as the key and a vector of nodes as the value to represent the partition.
Here is the problem:
assuming two persons are registered in a social networking website, how to decide whether they are connected or not?
my analysis (after reading more): actually, the question is looking for - the shortest path from A to B in a graph. I think both BFS and Dijkstra's Algorithms works here and time complexity is exactly the same (O(V+E)) because it is an unweighted graph, so we can't take advantage of the priority queue. So, a simple queue could resolve the problem. But, both of them doesnt resolve the problem that: find the path between them.
Bidrectrol should be a better solution at this point.
To find a path between the two, you should begin with a breadth first search. First find all neighbors of A, then find all neighbors of all neighbors of A, etc. Once B is hit, not only do you have a path from A to B, but you also have a shortest such path.
Dijkstra's algorithm rocks, and you may be able to speed this up by working from both end, i.e. find neighbors of A and neighbors of B, and compare.
If you do a depth first search, then you're following one path at a time. This will be much much slower.
If you do dfs for finding whether two people are connected on a social network, then it will take too long!
You already know the two persons, so you should use Bidirectional Search.. But, simple bidirectional search won't be enough for a graph as big as a social networking site. You will have to use some heuristics. Wikipedia page has some links to it.
You may also be able to use A* search. From wikipedia : "A* uses a best-first search and finds the least-cost path from a given initial node to one goal node (out of one or more possible goals)."
Edit: I suggest A* because "The additional complexity of performing a bidirectional search means that the A* search algorithm is often a better choice if we have a reasonable heuristic." So, if you can't form a reasonable heuristic, then use Bidirectional search. (Forming a good heuristic is never easy ;).)
One way is to use Union Find, add all links union(from,to), and if find(A) is find(B) is True then A and B are connected. This avoids the recursive search but it actually computes the connectivity of all pairs and doesn't give you the paths that connects A and B.
I think that the true criteria is: there are at least N paths between A and B shorter then K, or A and B are connected diectly. I would go with K = 3 and N near 5, i.e. have 5 common friends.
Note: answer edited.
Any method might end up being very slow. If you need to do this repeatedly, it's best to find the connected components of the graph, after which the task becomes a trivial O(1) operation: if two people are in the same component, they are connected.
Note that finding connected components for the first time might be slow, but keeping them updated as new edges/nodes are added to the graph is fast.
There are several methods for finding connected components.
One method is to construct the Laplacian of the graph, and look at its eigenvalues / eigenvectors. The number of zero eigenvalues gives you the number of connected components. The non-zero elements of the corresponding eigenvectors gives the nodes belonging to the respective components.
Another way is along the following lines:
Create a transformation table of nodes. Element n of the array contains the index of the node that node n transforms to.
Loop through all edges (i,j) in the graph (denoting a connection between i and j):
Compute recursively which node do i and j transform to based on the current table. Let us denote the results by k and l. Update entry k to make it transform to l. Update entries i and j to point to l as well.
Loop through the table again, and update each entry to point directly to the node it recursively transforms to.
Now nodes in the same connected component will have the same entry in the transformation table. So to check if two nodes are connected, just check if they transform to the same value.
Every time a new node or edge is added to the graph, the transformation table needs to be updated, but this update will be much faster than the original calculation of the table.
Is there an algorithm that can check, in a directed graph, if a vertex, let's say V2, is reachable from a vertex V1, without traversing all the vertices?
You might find a route to that node without traversing all the edges, and if so you can give a yes answer as soon as you do. Nothing short of traversing all the edges can confirm that the node isn't reachable (unless there's some other constraint you haven't stated that could be used to eliminate the possibility earlier).
Edit: I should add that it depends on how often you need to do queries versus how large (and dense) your graph is. If you need to do a huge number of queries on a relatively small graph, it may make sense to pre-process the data in the graph to produce a matrix with a bit at the intersection of any V1 and V2 to indicate whether there's a connection from V1 to V2. This doesn't avoid traversing the graph, but it can avoid traversing the graph at the time of the query. I.e., it's basically a greedy algorithm that assumes you're going to eventually use enough of the combinations that it's easiest to just traverse them all and store the result. Depending on the size of the graph, the pre-processing step may be slow, but once it's done executing a query becomes quite fast (constant time, and usually a pretty small constant at that).
Depth first search or breadth first search. Stop when you find one. But there's no way to tell there's none without going through every one, no. You can improve the performance sometimes with some heuristics, like if you have additional information about the graph. For example, if the graph represents a coordinate space like a real map, and most of the time you know that there's going to be a mostly direct path, then you can attempt to have the depth-first search look along lines that "aim towards the target". However, imagine the case where the start and end points are right next to each other, but with no vector inbetween, and to find it, you have to go way out of the way. You have to check every case in order to be exhaustive.
I doubt it has a name, but a breadth-first search might go like this:
Add V1 to a queue of nodes to be visited
While there are nodes in the queue:
If the node is V2, return true
Mark the node as visited
For every node at the end of an outgoing edge which is not yet visited:
Add this node to the queue
End for
End while
Return false
Create an adjacency matrix when the graph is created. At the same time you do this, create matrices consisting of the powers of the adjacency matrix up to the number of nodes in the graph. To find if there is a path from node u to node v, check the matrices (starting from M^1 and going to M^n) and examine the value at (u, v) in each matrix. If, for any of the matrices checked, that value is greater than zero, you can stop the check because there is indeed a connection. (This gives you even more information as well: the power tells you the number of steps between nodes, and the value tells you how many paths there are between nodes for that step number.)
(Note that if you know the number of steps in the longest path in your graph, for whatever reason, you only need to create a number of matrices up to that power. As well, if you want to save memory, you could just store the base adjacency matrix and create the others as you go along, but for large matrices that may take a fair amount of time if you aren't using an efficient method of doing the multiplications, whether from a library or written on your own.)
It would probably be easiest to just do a depth- or breadth-first search, though, as others have suggested, not only because they're comparatively easy to implement but also because you can generate the path between nodes as you go along. (Technically you'd be generating multiple paths and discarding loops/dead-end ones along the way, but whatever.)
In principle, you can't determine that a path exists without traversing some part of the graph, because the failure case (a path does not exist) cannot be determined without traversing the entire graph.
You MAY be able to improve your performance by searching backwards (search from destination to starting point), or by alternating between forward and backward search steps.
Any good AI textbook will talk at length about search techniques. Elaine Rich's book was good in this area. Amazon is your FRIEND.
You mentioned here that the graph represents a road network. If the graph is planar, you could use Thorup's Algorithm which creates an O(nlogn) space data structure that takes O(nlogn) time to build and answers queries in O(1) time.
Another approach to this problem would allow you to ignore all of the vertices. If you were to only look at the edges, you can produce a transitive closure array that will show you each vertex that is reachable from any other vertex.
Start with your list of edges:
Va -> Vc
Va -> Vd
....
Create an array with start location as the rows and end location as the columns. Fill the arrays with 0. For each edge in the list of edges, place a one in the start,end coordinate of the edge.
Now you iterate a few times until either V1,V2 is 1 or there are no changes.
For each row:
NextRowN = RowN
For each column that is true for RowN
Use boolean OR to OR in the results of that row of that number with the current NextRowN.
Set RowN to NextRowN
If you run this algorithm until the end, you will quickly have a complete list of all reachable vertices without looking at any of them. The runtime is proportional to the number of edges. This would work well with a reasonable implementation and a reasonable number of edges.
A slightly more complex version of this algorithm would be to only calculate the vertices reachable by V1. To do this, you would focus your scope on the ones that are currently reachable at any given time. You can also limit adding rows to only one time, since the other rows are never changing.
In order to be sure, you either have to find a path, or traverse all vertices that are reachable from V1 once.
I would recommend an implementation of depth first or breadth first search that stops when it encounters a vertex that it has already seen. The vertex will be processed on the first occurrence only. You need to make sure that the search starts at V1 and stops when it runs out of vertices or encounters V2.