Implementation (Training) - 2024.2 English - XD160

Vitis Libraries

Document ID
XD160
Release Date
2024-11-29
Version
2024.2 English

SGD Framework is basically two parts: gradient calculation and weight update. Gradient calculation loads data from DDR and calculates the gradient of the preset weight and is a dataflow function. Weight update calculates the next weight based on the gradient and the chosen method and is executed after gradient calculation. The block diagram is shown as follows.

3 stage dataflow

The correctness of Linear Regression/LASSO Regression/Ridge Regression Training using the SGD framework is verified by comparing results with Spark mllib. The results are identical.