with - Binary trees: Getting the path of an element from its signature



what is path in binary tree (1)

Assume you have a binary tree which classifies images. Each node is a different binary test. When an image is fed to the tree, a unique path through the tree is generated.

A path is described as a binary word which is as long as the depth of the tree.

For instance, for a 2-stage binary tree, an example of path would be (0,1) ((left,right) thus ending in the second leaf of the tree from the left).

We also assume that for any image being fed to the tree, all node tests are executable. Thus, we can define a signature for any image: this signature is a binary word which length is the number of nodes.

For instance, for a 2-stage binary tree, an example of signature would be s = {0,1,1}

s[i] is the binary result from the test of the ith node.

I am looking for a way to get the path of an image from its signature.

A naive way of doing it would be to jump from index to index of the signature with appropriate decreasing leap length.

However, I wonder if there could be a bitwise calculation which could yield the result (The indexes in the signature can be reorganized if necessary).

Is this possible? If yes, how?

https://code.i-harness.com


With 1 stage there is 1 test.

With 2 stages there are 1+2 tests.

With 3 stages there are 1+2+4 tests.

With 4 stages there are 1+2+4+8 tests.

With 5 stages there are 1+2+4+8+16=31 tests.

One approach to computing the path faster is to test the first bit, and then extract the corresponding set of 15bits that map to the next 4 stages. There are only 2**15=32768 different choices so the path can be looked up in a table.

Pseudocode:

 if (x&(1<<30))
    x >>= 15;
 path = lookup[x & 0x7fff];

lookup is a 2**15 length array that you can precalculate.