You are tasked to optimize a machine learning application on Hadoop. This application has a map stage and a reduce stage. Before optimization, the map stage accounts for 55% of the execution time, whereas the reduce stage accounts for 45%. For the map stage, you can optimize it to use a NUMA machine with 48 CPU cores, with an expected speedup of 1.1. For the reduce stage, you can optimize it to use GPU rocessors, with an expected speedup of 9. Limited by time, you can only choose one stage to optimize.Which stage should you choose? Please justify your answer with Amdahls Law. There will be no pointswithout justification.