Data update method and device
A data update and updated technology, applied in the computer field, can solve the problem of large data processing volume of the first server
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0023] The method for updating data provided in this embodiment may be, but is not limited to, applied to scenarios of topic model training. Most of the existing topic model training methods use a distributed architecture based on parameter servers to train topic models. like figure 1 As shown, the existing topic model training method stores the word-topic matrix in the server, and stores the document-topic matrix in the worker. In the training process of the topic model, each worker obtains the word-topic matrix from multiple servers to the local before each iteration, and then performs the Gibbs sampling algorithm. Since the Gibbs sampling algorithm will change the topic assignment of words, both the word-topic matrix and document-topic matrix need to be updated. Therefore, the worker will push the update of the word-topic to the server for updating. When the worker has sampled all the words in the machine, a round of iteration ends.
[0024] Because in each round of iter...
Embodiment 2
[0107] According to an embodiment of the present invention, there is also provided a data update device for implementing the above data update method, which is applied to the first server, such as Figure 6 As shown, the device includes:
[0108] 1) The first processing module 62 is configured to push the training instruction information to the second server, and pull the second part of the matrix corresponding to the first word set in the second matrix from the second server, wherein the training instruction information carries There is a second word set and a first partial matrix corresponding to the second word set in the first matrix, and the training instruction information is used to instruct the second server to update the first partial matrix and the second matrix, the first server stores the first matrix, the second server stores the second matrix;
[0109] 2) the first update module 64, for updating the second part matrix and the first matrix according to the second...
Embodiment 3
[0188] The application environment of the embodiment of the present invention may be, but not limited to, refer to the application environment in Embodiment 1, which will not be repeated in this embodiment. The embodiment of the present invention provides an optional specific application example for implementing the above data update method.
[0189] As an optional embodiment, the above data update method can be applied to but not limited to such as Figure 8 In the scenario shown where the matrix stored in the server is updated.
[0190] The data update scheme in this embodiment can be applied to machine learning tasks such as advertisement recommendation, text clustering, and user behavior analysis, but is not limited to. The topic model is a machine learning algorithm widely used in text analysis. This scheme provides A solution for efficiently training topic models in a modern environment. Users do not need the details of relational algorithm execution when using it, and...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More