priority_queue becomes extremely slow in debug mode - c++

I am currently writing an A* pathfinding algorithm for a game and came across a very strange performance problem regarding priority_queue's.
I am using a typical 'open nodes list', where I store found, but yet unprocessed nodes. This is implemented as an STL priority_queue (openList) of pointers to PathNodeRecord objects, which store information about a visited node. They are sorted by the estimated cost to get there (estimatedTotalCost).
Now I noticed that whenever the pathfinding method is called, the respective AI thread gets completely stuck and takes several (~5) seconds to process the algorithm and calculate the path. Subsequently I used the VS2013 profiler to see, why and where it was taking so long.
As it turns out, the pushing to and popping from the open list (the priority_queue) takes up a very large amount of time. I am no expert in STL containers, but I never had problems with their efficiency before and this is just weird to me.
The strange thing is that this only occurs while using VS's 'Debug' build configuration. The 'Release' conf. works fine for me and the times are back to normal.
Am I doing something fundamentally wrong here or why is the priority_queue performing so badly for me? The current situation is unacceptable to me, so if I cannot resolve it soon, I will need to fall back to using a simpler container and inserting it to the right place manually.
Any pointers to why this might be occuring would be very helpful!
.
Here is a snippet of what the profiler shows me:
http://i.stack.imgur.com/gEyD3.jpg
.
Code parts:
Here is the relevant part of the pathfinding algorithm, where it loops the open list until there are no open nodes:
// set up arrays and other variables
PathNodeRecord** records = new PathNodeRecord*[graph->getNodeAmount()]; // holds records for all nodes
std::priority_queue<PathNodeRecord*> openList; // holds records of open nodes, sorted by estimated rest cost (most promising node first)
// null all record pointers
memset(records, NULL, sizeof(PathNodeRecord*) * graph->getNodeAmount());
// set up record for start node and put into open list
PathNodeRecord* startNodeRecord = new PathNodeRecord();
startNodeRecord->node = startNode;
startNodeRecord->connection = NULL;
startNodeRecord->closed = false;
startNodeRecord->costToHere = 0.f;
startNodeRecord->estimatedTotalCost = heuristic->estimate(startNode, goalNode);
records[startNode] = startNodeRecord;
openList.push(startNodeRecord);
// ### pathfind algorithm ###
// declare current node variable
PathNodeRecord* currentNode = NULL;
// loop-process open nodes
while (openList.size() > 0) // while there are open nodes to process
{
// retrieve most promising node and immediately remove from open list
currentNode = openList.top();
openList.pop(); // ### THIS IS, WHERE IT GETS STUCK
// if current node is the goal node, end the search here
if (currentNode->node == goalNode)
break;
// look at connections outgoing from this node
for (auto connection : graph->getConnections(currentNode->node))
{
// get end node
PathNodeRecord* toNodeRecord = records[connection->toNode];
if (toNodeRecord == NULL) // UNVISITED -> path record needs to be created and put into open list
{
// set up path node record
toNodeRecord = new PathNodeRecord();
toNodeRecord->node = connection->toNode;
toNodeRecord->connection = connection;
toNodeRecord->closed = false;
toNodeRecord->costToHere = currentNode->costToHere + connection->cost;
toNodeRecord->estimatedTotalCost = toNodeRecord->costToHere + heuristic->estimate(connection->toNode, goalNode);
// store in record array
records[connection->toNode] = toNodeRecord;
// put into open list for future processing
openList.push(toNodeRecord);
}
else if (!toNodeRecord->closed) // OPEN -> evaluate new cost to here and, if better, update open list entry; otherwise skip
{
float newCostToHere = currentNode->costToHere + connection->cost;
if (newCostToHere < toNodeRecord->costToHere)
{
// update record
toNodeRecord->connection = connection;
toNodeRecord->estimatedTotalCost = newCostToHere + (toNodeRecord->estimatedTotalCost - toNodeRecord->costToHere);
toNodeRecord->costToHere = newCostToHere;
}
}
else // CLOSED -> evaluate new cost to here and, if better, put back on open list and reset closed status; otherwise skip
{
float newCostToHere = currentNode->costToHere + connection->cost;
if (newCostToHere < toNodeRecord->costToHere)
{
// update record
toNodeRecord->connection = connection;
toNodeRecord->estimatedTotalCost = newCostToHere + (toNodeRecord->estimatedTotalCost - toNodeRecord->costToHere);
toNodeRecord->costToHere = newCostToHere;
// reset node to open and push into open list
toNodeRecord->closed = false;
openList.push(toNodeRecord); // ### THIS IS, WHERE IT GETS STUCK
}
}
}
// set node to closed
currentNode->closed = true;
}
Here is my PathNodeRecord with the 'less' operator overloading to enable sorting in priority_queue:
namespace AI
{
struct PathNodeRecord
{
Node node;
NodeConnection* connection;
float costToHere;
float estimatedTotalCost;
bool closed;
// overload less operator comparing estimated total cost; used by priority queue
// nodes with a higher estimated total cost are considered "less"
bool operator < (const PathNodeRecord &otherRecord)
{
return this->estimatedTotalCost > otherRecord.estimatedTotalCost;
}
};
}

std::priority_queue<PathNodeRecord*> openList
I think the reason is that you have a priority_queue of pointers to PathNodeRecord.
and there is no ordering defined for the pointers.
try changing it to std::priority_queue<PathNodeRecord> first, if it makes a difference then all you need is passing on your own comparator that knows how to compare pointers to PathNodeRecord, it will just dereference the pointers first and then do the comparison.
EDIT:
taking a wild guess about why did you get an extremely slow execution time, I think the pointers were compared based on their address. and the addresses were allocated starting from one point in memory and going up.
and so that resulted in the extreme case of your heap (the heap as in data structure not the memory part), so your heap was actually a list, (a tree where each node had one children node and so on).
and so you operation took a linear time, again just a guess.

You cannot expect a debug build to be as fast as a release optimized one, but you seems to do a lot of dynamic allocation that may interact badly with the debug runtime.
I suggest you to add _NO_DEBUG_HEAP=1 in the environment setting of the debug property page of your project.

Related

Segmentation fault in recursive function when using smart pointers

I get a segmentation fault in the call to
auto n1=std::make_shared<Node>(n,n->x+i,n->y+j);
after a few recursive calls. Strange thing is that it's always at the same point in time. Can anyone spot the problem?
This is an implementation for a dynamic programming problem and here I'm accumulating the costs of a path. I have simplified the cost function but in this example the problem still occurs.
void HorizonLineDetector::dp(std::shared_ptr<Node> n)
{
n->cost= 1 + n->prev->cost;
//Check if we reached the last column(done!)
if (n->x==current_edges.cols-1)
{
//Save the info in the last node if it's the cheapest path
if (last_node->cost > n->cost)
{
last_node->cost=n->cost;
last_node->prev=n;
}
}
else
{
//Check for neighboring pixels to see if they are edges, launch dp with all the ones that are
for (int i=0;i<2;i++)
{
for (int j=-1;j<2;j++)
{
if (i==0 && j==0) continue;
if (n->x+i >= current_edges.cols || n->x+i < 0 ||
n->y+j >= current_edges.rows || n->y+j < 0) continue;
if (current_edges.at<char>(n->y+j,n->x+i)!=0)
{
auto n1=std::make_shared<Node>(n,n->x+i,n->y+j);
//n->next.push_back(n1);
nlist.push_back(n1);
dp(n1);
}
}
}
}
}
class Node
{
public:
Node(){}
Node(std::shared_ptr<Node> p,int x_,int y_){prev=p;x=x_;y=y_;lost=0;}
Node(Node &n1){x=n1.x;y=n1.y;cost=n1.cost;lost=n1.lost;prev=n1.prev;}//next=n1.next;}
std::shared_ptr<Node> prev; //Previous and next nodes
int cost; //Total cost until now
int lost; //Number of steps taken without a clear path
int x,y;
Node& operator=(const Node &n1){x=n1.x;y=n1.y;cost=n1.cost;lost=n1.lost;prev=n1.prev;}//next=n1.next;}
Node& operator=(Node &&n1){x=n1.x;y=n1.y;cost=n1.cost;lost=n1.lost;prev=n1.prev;n1.prev=nullptr;}//next=n1.next;n1.next.clear();}
};
Your code looks like a pathological path search, in that it checks almost every path and doesn't keep track of paths it has already checked you can get to more than one way.
This will build recursive depth equal to the length of the longest path, and then the next longest path, and ... down to the shortest one. Ie, something like O(# of pixels) depth.
This is bad. And, as call stack depth is limited, will crash you.
The easy solution is to modify dp into dp_internal, and have dp_internal return a vector of nodes to process next. Then write dp, which calls dp_internal and repeats on its return value.
std::vector<std::shared_ptr<Node>>
HorizonLineDetector::dp_internal(std::shared_ptr<Node> n)
{
std::vector<std::shared_ptr<Node>> retval;
...
if (current_edges.at<char>(n->y+j,n->x+i)!=0)
{
auto n1=std::make_shared<Node>(n,n->x+i,n->y+j);
//n->next.push_back(n1);
nlist.push_back(n1);
retval.push_back(n1);
}
...
return retval;
}
then dp becomes:
void HorizonLineDetector::dp(std::shared_ptr<Node> n)
{
std::vector<std::shared_ptr<Node>> nodes={n};
while (!nodes.empty()) {
auto node = nodes.back();
nodes.pop_back();
auto new_nodes = dp_internal(node);
nodes.insert(nodes.end(), new_nodes.begin(), new_nodes.end());
}
}
but (A) this will probably just crash when the number of queued-up nodes gets ridiculously large, and (B) this just patches over the recursion-causes-crash, doesn't make your algorithm suck less.
Use A*.
This involves keeping track of which nodes you have visited and what nodes to process next with their current path cost.
You then use heuristics to figure out which of the ones to process next you should check first. If you are on a grid of some sort, the heuristic is to use the shortest possible distance if nothing was in the way.
Add the cost to get to the node-to-process, plus the heuristic distance from that node to the destination. Find the node-to-process that has the least total. Process that one: you mark it as visited, and add all of its adjacent nodes to the list of nodes to process.
Never add a node to the list of nodes to process that you have already visited (as that is redundant work).
Once you have a solution, prune the list of nodes to process against any node whose current path value is greater than or equal to your solution. If you know your heuristic is a strong one (that it is impossible to get to the destination faster), you can even prune based off of the total of heuristic and current cost. Similarly, don't add to the list of nodes to process if it would be pruned by this paragraph.
The result is that your algorithm searches in a relatively strait line towards the target, and then expands outwards trying to find a way around any barriers. If there is a relatively direct route, it is used and the rest of the universe isn't even touched.
There are many optimizations on A* you can do, and even alternative solutions that don't rely on heuristics. But start with A*.

Is it possible to make efficient pointer-based binary heap implementations?

Is it even possible to implement a binary heap using pointers rather than an array? I have searched around the internet (including SO) and no answer can be found.
The main problem here is that, how do you keep track of the last pointer? When you insert X into the heap, you place X at the last pointer and then bubble it up. Now, where does the last pointer point to?
And also, what happens when you want to remove the root? You exchange the root with the last element, and then bubble the new root down. Now, how do you know what's the new "last element" that you need when you remove root again?
Solution 1: Maintain a pointer to the last node
In this approach a pointer to the last node is maintained, and parent pointers are required.
When inserting, starting at the last node navigate to the node below which a new last node will be inserted. Insert the new node and remember it as the last node. Move it up the heap as needed.
When removing, starting at the last node navigate to the second-to-last node. Remove the original last node and remember the the new last node just found. Move the original last node into the place of the deleted node and then move it up or down the heap as needed.
It is possible to navigate to the mentioned nodes in O(log(n)) time and O(1) space. Here is a description of the algorithms but the code is available below:
For insert: If the last node is a left child, proceed with inserting the new node as the right child of the parent. Otherwise... Start at the last node. Move up as long as the current node is a right child. If the root was not reached, move to the sibling node at the right (which necessarily exists). Then (whether or not the root was reached), move down to the left as long as possible. Proceed by inserting the new node as the left child of the current node.
For remove: If the last node is the root, proceed by removing the root. Otherwise... Start at the last node. Move up as long as the current node is a left child. If the root was not reached, move to the sibling left node (which necessarily exists). Then (whether or not the root was reached), move down to the right as long as possible. We have arrived at the second-to-last node.
However, there are some things to be careful about:
When removing, there are two special cases: when the last node is being removed (unlink the node and change the last node pointer), and when the second-to-last node is being removed (not really special but the possibility must be considered when replacing the deleted node with the last node).
When moving nodes up or down the heap, if the move affects the last node, the last-node pointer must be corrected.
Long ago I have made an implementation of this. In case it helps someone, here is the code. Algorithmically it should be correct (has also been subjected to stress testing with verification), but there is no warranty of course.
Solution 2: Reach the last node from the root
This solution requires maintaining the node count (but not parent pointers or the last node). The last (or second-to-last) node is found by navigating from the root towards it.
Assume the nodes are numbered starting from 1, as per the typical notation for binary heaps. Pick any valid node number and represent it in binary. Ignore the first (most significant) 1 bit. The remaining bits define the path from the root to that node; zero means left and one means right.
For example, to reach node 11 (=1011b), start at the root then go left (0), right (1), right (1).
This algorithm can be used in insert to find where to place the new node (follow the path for node node_count+1), and in remove to find the second-to-last-node (follow the path for node node_count-1).
This approach is used in libuv for timer management; see their implementation of the binary heap.
Usefulness of Pointer-based Binary Heaps
Many answers here and even literature say that an array-based implementation of a binary heap is strictly superior. However I contest that because there are situations where the use of an array is undesirable, typically because the upper size of the array is not known in advance and on-demand reallocations of an array are not deemed acceptable, for example due to latency or possibility of allocation failure.
The fact that libuv (a widely used event loop library) uses a binary heap with pointers only further speaks for this.
It is worth noting that the Linux kernel uses (pointer-based) red-black trees as a priority queue in a few cases, for example for CPU scheduling and timer management (for the same purpose as in libuv). I find it likely that changing these to use a pointer-based binary heap will improve performance.
Hybrid Approach
It is possible to combine Solution 1 and Solution 2 into a hybrid approach which dynamically picks either of the algorithms (for finding the last or second-to-last node), the one with a lower cost, measured in the number of edges that need to be traversed. Assume we want to navigate to node number N, and highest_bit(X) means the 0-based index of the highest-order bit in N (0 means the LSB).
The cost of navigating from the root (Solution 2) is highest_bit(N).
The cost of navigating from the previous node which is on the same level (Solution 1) is: 2 * (1 + highest_bit((N-1) xor N)).
Note that in the case of a level change the second equation will yield a wrong (too large) result, but in that case traversal from the root is more efficient anyway (for which the estimate is correct) and will be chosen, so there is no need for special handling.
Some CPUs have an instruction for highest_bit allowing very efficient implementation of these estimates. An alternative approach is to maintain the highest bit as a bit mask and do these calculation with bit masks instead of bit indices. Consider for example that 1 followed by N zeroes squared is equal to 1 followed by 2N zeroes).
In my testing it has turned out that Solution 1 is on average faster than Solution 2, and the hybrid approach appeared to have about the same average performance as Solution 2. Therefore the hybrid approach is only useful if one needs to minimize the worst-case time, which is (twice) better in Solution 2; since Solution 1 will in the worst case traverse the entire height of the tree up and then down.
Code for Solution 1
Note that the traversal code in insert is slightly different from the algorithm described above but still correct.
struct Node {
Node *parent;
Node *link[2];
};
struct Heap {
Node *root;
Node *last;
};
void init (Heap *h)
{
h->root = NULL;
h->last = NULL;
}
void insert (Heap *h, Node *node)
{
// If the heap is empty, insert root node.
if (h->root == NULL) {
h->root = node;
h->last = node;
node->parent = NULL;
node->link[0] = NULL;
node->link[1] = NULL;
return;
}
// We will be finding the node to insert below.
// Start with the current last node and move up as long as the
// parent exists and the current node is its right child.
Node *cur = h->last;
while (cur->parent != NULL && cur == cur->parent->link[1]) {
cur = cur->parent;
}
if (cur->parent != NULL) {
if (cur->parent->link[1] != NULL) {
// The parent has a right child. Attach the new node to
// the leftmost node of the parent's right subtree.
cur = cur->parent->link[1];
while (cur->link[0] != NULL) {
cur = cur->link[0];
}
} else {
// The parent has no right child. This can only happen when
// the last node is a right child. The new node can become
// the right child.
cur = cur->parent;
}
} else {
// We have reached the root. The new node will be at a new level,
// the left child of the current leftmost node.
while (cur->link[0] != NULL) {
cur = cur->link[0];
}
}
// This is the node below which we will insert. It has either no
// children or only a left child.
assert(cur->link[1] == NULL);
// Insert the new node, which becomes the new last node.
h->last = node;
cur->link[cur->link[0] != NULL] = node;
node->parent = cur;
node->link[0] = NULL;
node->link[1] = NULL;
// Restore the heap property.
while (node->parent != NULL && value(node->parent) > value(node)) {
move_one_up(h, node);
}
}
void remove (Heap *h, Node *node)
{
// If this is the only node left, remove it.
if (node->parent == NULL && node->link[0] == NULL && node->link[1] == NULL) {
h->root = NULL;
h->last = NULL;
return;
}
// Locate the node before the last node.
Node *cur = h->last;
while (cur->parent != NULL && cur == cur->parent->link[0]) {
cur = cur->parent;
}
if (cur->parent != NULL) {
assert(cur->parent->link[0] != NULL);
cur = cur->parent->link[0];
}
while (cur->link[1] != NULL) {
cur = cur->link[1];
}
// Disconnect the last node.
assert(h->last->parent != NULL);
h->last->parent->link[h->last == h->last->parent->link[1]] = NULL;
if (node == h->last) {
// Deleting last, set new last.
h->last = cur;
} else {
// Not deleting last, move last to node's place.
Node *srcnode = h->last;
replace_node(h, node, srcnode);
// Set new last unless node=cur; in this case it stays the same.
if (node != cur) {
h->last = cur;
}
// Restore the heap property.
if (srcnode->parent != NULL && value(srcnode) < value(srcnode->parent)) {
do {
move_one_up(h, srcnode);
} while (srcnode->parent != NULL && value(srcnode) < value(srcnode->parent));
} else {
while (srcnode->link[0] != NULL || srcnode->link[1] != NULL) {
bool side = srcnode->link[1] != NULL && value(srcnode->link[0]) >= value(srcnode->link[1]);
if (value(srcnode) > value(srcnode->link[side])) {
move_one_up(h, srcnode->link[side]);
} else {
break;
}
}
}
}
}
Two other functions are used: move_one_up moves a node one step up in the heap, and replace_node replaces moves an existing node (srcnode) into the place held by the node being deleted. Both work only by adjusting the links to and from the other nodes, there is no actual moving of data involved. These functions should not be hard to implement, and the mentioned link includes my implementations.
The pointer based implementation of the binary heap is incredibly difficult when compared to the array based implementation. But it is fun to code it. The basic idea is that of a binary tree. But the biggest challenge you will have is to keep it left-filled. You will have difficulty in finding the exact location as to where you must insert a node.
For that, you must know binary traversal. What we do is. Suppose our heap size is 6. We will take the number + 1, and convert it to bits. The binary representation of 7 is, "111". Now, remember to always omit the first bit. So, now we are left with "11". Read from left-to-right. The bit is '1', so, go to the right child of the root node. Then the string left is "1", the first bit is '1'. As you have only 1 bit left, this single bit tells you where to insert the new node. As it is '1' the new node must be the right child of the current node. So, the raw working of the process is that, convert the size of the heap into bits. Omit the first bit. According to the leftmost bit, go to the right child of the current node if it is '1', and to the left child of the current node if it is '0'.
After inserting the new node, you will bubble it up the heap. This tells you that you will be needing the parent pointer. So, you go once down the tree and once up the tree. So, your insertion operation will take O(log N).
As for the deletion, it is still a challenge to find the last node. I hope you are familiar with deletion in a heap where we swap it with the last node and do a heapify. But for that you need the last node, for that too, we use the same technique as we did for finding the location to insert the new node, but with a little twist. If you want to find the location of the last node, you must use the binary representation of the value HeapSize itself, not HeapSize + 1. This will take you to the last node. So, the deletion will also cost you O(log N).
I'm having trouble in posting the source code here, but you can refer to my blog for the source code. In the code, there is Heap Sort too. It is very simple. We just keep deleting the root node. Refer to my blog for explanation with figures. But I guess this explanation would do.
I hope my answer has helped you. If it did, let me know...! ☺
For those saying this is a useless exercise, there are a couple of (admittedly rare) use cases where a pointer-based solution is better. If the max size of the heap is unknown, then an array implementation will need to stop-and-copy into fresh storage when the array fills. In a system (e.g. embedded) where there are fixed response time constraints and/or where free memory exists, but not a big enough contiguous block, this may be not be acceptable. The pointer tree lets you allocate incrementally in small, fixed-size chunks, so it doesn't have these problems.
To answer the OP's question, parent pointers and/or elaborate tracking aren't necessary to determine where to insert the next node or find the current last one. You only need the bits in the binary rep of the heap's size to determine the left and right child pointers to follow.
Edit Just saw Vamsi Sangam#'s explanation of this algorithm. Nonetheless, here's a demo in code:
#include <stdio.h>
#include <stdlib.h>
typedef struct node_s {
struct node_s *lft, *rgt;
int data;
} NODE;
typedef struct heap_s {
NODE *root;
size_t size;
} HEAP;
// Add a new node at the last position of a complete binary tree.
void add(HEAP *heap, NODE *node) {
size_t mask = 0;
size_t size = ++heap->size;
// Initialize the mask to the high-order 1 of the size.
for (size_t x = size; x; x &= x - 1) mask = x;
NODE **pp = &heap->root;
// Advance pp right or left depending on size bits.
while (mask >>= 1) pp = (size & mask) ? &(*pp)->rgt : &(*pp)->lft;
*pp = node;
}
void print(NODE *p, int indent) {
if (!p) return;
for (int i = 0; i < indent; i++) printf(" ");
printf("%d\n", p->data);
print(p->lft, indent + 1);
print(p->rgt, indent + 1);
}
int main(void) {
HEAP h[1] = { NULL, 0 };
for (int i = 0; i < 16; i++) {
NODE *p = malloc(sizeof *p);
p->lft = p->rgt = NULL;
p->data = i;
add(h, p);
}
print(h->root, 0);
}
As you'd hope, it prints:
0
1
3
7
15
8
4
9
10
2
5
11
12
6
13
14
Sift-down can use the same kind of iteration. It's also possible to implement the sift-up without parent pointers using either recursion or an explicit stack to "save" the nodes in the path from root to the node to be sifted.
A binary heap is a complete binary tree obeying the heap property. That's all. The fact that it can be stored using an array, is just nice and convenient. But sure, you can implement it using a linked structure. It's a fun exercise! As such, it is mostly useful as an exercise or in more advanced datastructures( meldable, addressable priority queues for example ), as it is quite a bit more involved than doing the array version. For example, think about siftup/siftdown procedures, and all the edge cutting/sewing you'll need to get right. Anyways, it's not too hard, and once again, good fun!
There are a number of comments pointing out that by a strict definition it is possible to implement a binary heap as a tree and still call it a binary heap.
Here is the problem -- there is never a reason to do so since using an array is better in every way.
If you do searches to try to find information on how to work with a heap using pointers you are not going to find any -- no one bothers since there is no reason to implement a binary heap in this way.
If you do searches on trees you will find lots of helpful materials. This was the point of my original answer. There is nothing that stops people from doing it this way but there is never a reason to do so.
You say -- I have to do so, I've got an legacy system and I have pointers to nodes I need to put them in a heap.
Make an array of those pointers and work with them in this array as you would a standard array based heap, when you need the contents dereference them. This will work better than any other way of implementing your system.
I can think of no other reason to implement a heap using pointers.
Original Answer:
If you implement it with pointers then it is a tree. A heap is a heap because of how you can calculate the location of the children as a location in the array (2 * node index +1 and 2 * node index + 2).
So no, you can't implement it with pointers, if you do you've implemented a tree.
Implementing trees is well documented if you search you will find your answers.
I have searched around the internet (including SO) and no answer can be found.
Funny, because I found an answer on SO within moments of googling it. (Same Google search led me here.)
Basically:
The node should have pointers to its parent, left child, and right child.
You need to keep pointers to:
the root of the tree (root) (duh)
the last node inserted (lastNode)
the leftmost node of the lowest level (leftmostNode)
the rightmost node of the next-to-lowest level (rightmostNode)
Now, let the node to be inserted be nodeToInsert. Insertion algorithm in pseudocode:
void insertNode(Data data) {
Node* parentNode, nodeToInsert = new Node(data);
if(root == NULL) { // empty tree
parent = NULL;
root = nodeToInsert;
leftmostNode = root;
rightmostNode = NULL;
} else if(lastNode.parent == rightmostNode && lastNode.isRightChild()) {
// level full
parentNode = leftmostNode;
leftmostNode = nodeToInsert;
parentNode->leftChild = nodeToInsert;
rightmostNode = lastNode;
} else if (lastNode.isLeftChild()) {
parentNode = lastNode->parent;
parentNode->rightChild = nodeToInsert;
} else if(lastNode.isRightChild()) {
parentNode = lastNode->parent->parent->rightChild;
parentNode->leftChild = nodeToInsert;
}
nodeToInsert->parent = parentNode;
lastNode = nodeToInsert;
heapifyUp(nodeToInsert);
}
Pseudocode for deletion:
Data deleteNode() {
Data result = root->data;
if(root == NULL) throw new EmptyHeapException();
if(lastNode == root) { // the root is the only node
free(root);
root = NULL;
} else {
Node* newRoot = lastNode;
if(lastNode == leftmostNode) {
newRoot->parent->leftChild = NULL;
lastNode = rightmostNode;
rightmostNode = rightmostNode->parent;
} else if(lastNode.isRightChild()) {
newRoot->parent->rightChild = NULL;
lastNode = newRoot->parent->leftChild;
} else if(lastNode.isLeftChild()) {
newRoot->parent->leftChild = NULL;
lastNode = newRoot->parent->parent->leftChild->rightChild;
}
newRoot->leftChild = root->leftChild;
newRoot->rightChild = root->rightChild;
newRoot->parent = NULL;
free(root);
root = newRoot;
heapifyDown(root);
}
return result;
}
heapifyUp() and heapifyDown() shouldn’t be too hard, though of course you’ll have to make sure those functions don’t make leftmostNode, rightmostNode, or lastNode point at the wrong place.
TL;DR Just use a goddamn array.

Low Memory Shortest Path Algorithm

I have a global unique path table which can be thought of as a directed un-weighted graph. Each node represents either a piece of physical hardware which is being controlled, or a unique location in the system. The table contains the following for each node:
A unique path ID (int)
Type of component (char - 'A' or 'L')
String which contains a comma separated list of path ID's which that node is connected to (char[])
I need to create a function which given a starting and ending node, finds the shortest path between the two nodes. Normally this is a pretty simple problem, but here is the issue I am having. I have a very limited amount of memory/resources, so I cannot use any dynamic memory allocation (ie a queue/linked list). It would also be nice if it wasn't recursive (but it wouldn't be too big of an issue if it was as the table/graph itself if really small. Currently it has 26 nodes, 8 of which will never be hit. At worst case there would be about 40 nodes total).
I started putting something together, but it doesn't always find the shortest path. The pseudo code is below:
bool shortestPath(int start, int end)
if start == end
if pathTable[start].nodeType == 'A'
Turn on part
end if
return true
else
mark the current node
bool val
for each node in connectedNodes
if node is not marked
val = shortestPath(node.PathID, end)
end if
end for
if val == true
if pathTable[start].nodeType == 'A'
turn on part
end if
return true
end if
end if
return false
end function
Anyone have any ideas how to either fix this code, or know something else that I could use to make it work?
----------------- EDIT -----------------
Taking Aasmund's advice, I looked into implementing a Breadth First Search. Below I have some c# code which I quickly threw together using some pseudo code I found online.
pseudo code found online:
Input: A graph G and a root v of G
procedure BFS(G,v):
create a queue Q
enqueue v onto Q
mark v
while Q is not empty:
t ← Q.dequeue()
if t is what we are looking for:
return t
for all edges e in G.adjacentEdges(t) do
u ← G.adjacentVertex(t,e)
if u is not marked:
mark u
enqueue u onto Q
return none
C# code which I wrote using this code:
public static bool newCheckPath(int source, int dest)
{
Queue<PathRecord> Q = new Queue<PathRecord>();
Q.Enqueue(pathTable[source]);
pathTable[source].markVisited();
while (Q.Count != 0)
{
PathRecord t = Q.Dequeue();
if (t.pathID == pathTable[dest].pathID)
{
return true;
}
else
{
string connectedPaths = pathTable[t.pathID].connectedPathID;
for (int x = 0; x < connectedPaths.Length && connectedPaths != "00"; x = x + 3)
{
int nextNode = Convert.ToInt32(connectedPaths.Substring(x, 2));
PathRecord u = pathTable[nextNode];
if (!u.wasVisited())
{
u.markVisited();
Q.Enqueue(u);
}
}
}
}
return false;
}
This code runs just fine, however, it only tells me if a path exists. That doesn't really work for me. Ideally what I would like to do is in the block "if (t.pathID == pathTable[dest].pathID)" I would like to have either a list or a way to see what nodes I had to pass through to get from the source and destination, such that I can process those nodes there, rather than returning a list to process elsewhere. Any ideas on how i could make that change?
The most effective solution, if you're willing to use static memory allocation (or automatic, as I seem to recall that the C++ term is), is to declare a fixed-size int array (of size 41, if you're absolutely certain that the number of nodes will never exceed 40). By using two indices to indicate the start and end of the queue, you can use this array as a ring buffer, which can act as the queue in a breadth-first search.
Alternatively: Since the number of nodes is so small, Bellman-Ford may be fast enough. The algorithm is simple to implement, does not use recursion, and the required extra memory is only a distance (int, or even byte in your case) and a predecessor id (int) per node. The running time is O(VE), alternatively O(V^3), where V is the number of nodes and E is the number of edges.

Kruskal’s Minimum Spanning Tree Algorithm (C++)

I'm working though an assignment on the Stanford CS106B C++ course, but I'm massively stuck implementing Kruskal's algorithm to find a minimum spanning tree.
To be more specific, I can't figure out the logic to determine whether to add an arc/ vertex to the tree. These are the instructions I've been given:
"The strategy you will use is based on tracking connected sets. For each node, maintain the set of the nodes that are connected to it. At
the start, each node is connected only to itself. When a new arc is
added, you merge the sets of the two endpoints into one larger combined set that both nodes are now connected to. When considering an
arc, if its two endpoints already belong to the same connected set,
there is no point in adding that arc and thus you skip it."
void getMinSpanTree(graphT *&graph)
{
Map<Set <nodeT *> > connections;
// Create set of arcs in decreasing order
Set<arcT *> arcs(costCmp);
Set<arcT *>::Iterator gItr = graph->arcs.iterator();
while (gItr.hasNext()) {
arcT *arc = gItr.next();
arcs.add(arc);
}
// Initialise map with initial node connections
Set<nodeT *>::Iterator nItr = graph->nodes.iterator();
while (nItr.hasNext()) {
nodeT *node = nItr.next();
Set<nodeT *> nodes;
nodes.add(node);
connections.add(node->name, nodes);
}
// Iterate through arcs
Set<arcT *>::Iterator aItr = arcs.iterator();
while (aItr.hasNext()) {
arcT *arc = aItr.next();
if (connections[arc->start->name].equals(connections[arc->finish->name])) {
Set<nodeT *> nodes = connections[arc->start->name];
nodes.unionWith(connections[arc->finish->name]);
connections[arc->start->name] = nodes;
connections[arc->finish->name] = nodes;
// Update display with arc
coordT start = {arc->start->x, arc->start->y};
coordT finish = {arc->finish->x, arc->finish->y};
DrawLineBetween(start, finish, HIGHLIGHT_COLOR);
}
}
}
I know the line:
if (connections[arc->start->name].equals(connections[arc->finish->name])) {
needs to be changed. Does anyone know what it should be? :)
One simple solution would be to iterate over nodes in
connections[arc->start->name]
and see if they match
arc->finish->name
If so, arc->start->name and arc->finish->name are connected and there's no point in merging the two sets.

Improving my quadtree design?

I have an application which is used for displaying and modifying huge volumes of point cloud data from lidar files (up to few gigabytes each, sometimes loaded in simultaneously). In the app the user is able to view a 2D image of loaded points (from the top) and select a profile to view in another window (from the side). Again this involves millions of points and they are displayed using OpenGL.
To handle the data there is also a quadtree library, which works, but is extremely slow. It has been used for some time, but recently the lidar point format changed and the LidarPoint object needed a number of attributes (class members) added, which cause it to grow in size in turn affecting the performance to almost unusable level (think 5 minutes to load a single 2GB file).
The quadtree currently consist of pointers to PointBucket objects which are simply arrays of LidarPoint objects with specified capacity and defined boundaries (for spatial queries). If the bucket capacity is exceeded it splits into four buckets. There is also kind of a caching system in place which causes point buckets to get dumped to disk when the point data is taking too much memory. These are then loaded back into memory if needed. Finally every PointBucket contains subbuckets/resolution levels which hold every n-th point of the original data and are used when displaying the data depending on the zoom level. That is because displaying few million points at once, while that level of detail is not necessary, is just extremely slow.
I hope you can get a picture from this. If not please ask and I can provide some more details or upload more code. For example here is the current (and slow) insert method:
// Insert in QuadTree
bool QuadtreeNode::insert(LidarPoint newPoint)
{
// if the point dosen't belong in this subset of the tree return false
if (newPoint.getX() < minX_ || newPoint.getX() > maxX_ ||
newPoint.getY() < minY_ || newPoint.getY() > maxY_)
{
return false;
}
else
{
// if the node has overflowed and is a leaf
if ((numberOfPoints_ + 1) > capacity_ && leaf_ == true)
{
splitNode();
// insert the new point that caused the overflow
if (a_->insert(newPoint))
{
return true;
}
if (b_->insert(newPoint))
{
return true;
}
if (c_->insert(newPoint))
{
return true;
}
if (d_->insert(newPoint))
{
return true;
}
throw OutOfBoundsException("failed to insert new point into any \
of the four child nodes, big problem");
}
// if the node falls within the boundary but this node not a leaf
if (leaf_ == false)
{
return false;
}
// if the node falls within the boundary and will not cause an overflow
else
{
// insert new point
if (bucket_ == NULL)
{
bucket_ = new PointBucket(capacity_, minX_, minY_, maxX_, maxY_,
MCP_, instanceDirectory_, resolutionBase_,
numberOfResolutionLevels_);
}
bucket_->setPoint(newPoint);
numberOfPoints_++;
return true;
}
}
}
// Insert in PointBucket (quadtree holds pointers to PointBuckets which hold the points)
void PointBucket::setPoint(LidarPoint& newPoint)
{
//for each sub bucket
for (int k = 0; k < numberOfResolutionLevels_; ++k)
{
// check if the point falls into this subbucket (always falls into the big one)
if (((numberOfPoints_[0] + 1) % int(pow(resolutionBase_, k)) == 0))
{
if (!incache_[k])
cache(true, k);
// Update max/min intensity/Z values for the bucket.
if (newPoint.getIntensity() > maxIntensity_)
maxIntensity_ = newPoint.getIntensity();
else if (newPoint.getIntensity() < minIntensity_)
minIntensity_ = newPoint.getIntensity();
if (newPoint.getZ() > maxZ_)
maxZ_ = newPoint.getZ();
else if (newPoint.getZ() < minZ_)
minZ_ = newPoint.getZ();
points_[k][numberOfPoints_[k]] = newPoint;
numberOfPoints_[k]++;
}
}
}
Now my question is if you can think of a way to improve this design? What are some general strategies when dealing with huge amounts of data that doesn't fit into memory? How can I make the quadtree more efficient? Is there a way to speed up rendering of points?
Now my question is if you can think of a way to improve this design?
Yes: Don't store the objects itself in the quadtree. Put them into a flat structure (array, linked list, etc.) and have the Quadtree just keep a pointer to the actual objects. If the quadtree has a certain depth (on all nodes), you could flatten it as well.