This document describes the structure and execution of svm prediction, implemented as a svmPredict function.
Svm_predict is a function to predict a sample’s classification by trained weight vector and sample feature vector. This function provide a stream in, stream out module to easily get the prediction class of a sample.
The structure of svmPredict is described as follows. The primitive have two function which are getWeight and getPredictions. The input of getWeight is stream, and this part stores the weight vector in RAM. getPredictions has a dataflow region including two functions. One is dot_multiply, and the other is tag type transform.
The hardware resource utilization of svm_prediction is shown in the following table (work as 300 MHz).
| LUT | FF | BRAM | URAM | DSP |
| 27114 | 37335 | 26 | 0 | 133 |