Machine learning sample learning more discussion research, record the usual idea and so on, welcome to participate in the discussion??

CodePudding user response:

Idea 1:
is based on training package all instances in the build tree structure
Steps:
1:1) the concept of training are all instances in the package, for instance pool,
?? 2) are instances from the pool of random an instance as the root node;
?? 3) to calculate the rest of the instance and similarity of the root node, such as the cosine similarity;
?? 4) is less than a certain threshold, such as less than 0.5 0.5 0.5 instance as left subtree, the rest is right subtree;
?? 5) repeat the above steps, each node needs to record the package's serial number;
?? 6) for each package, mapping, in each instance, starting from the root node similarity walk path, use Huffman encoding record, for example,
?? 7) the question now is how to unified the length of the mapping?
Participants: ji, the day I like elder brother

CodePudding user response:

CodePudding user response:

This lazy time overhead is larger

CodePudding user response:

How to done to reduce time costs, can be selected according to the proportion, if is a full binary tree, you need to control the number of instances in 2 ^ n - 1; So on subsequent package mapping can also control the dimension of the vector

CodePudding user response:

Is surely worth a try ha ha ha