I am using a numpy based library for which the first step is to vectorize the array of input elements. It takes 0.7 seconds when the debugger is off and 33 seconds with the debugger on. For more complex operations, this ratio is still maintained - and therefore low tens of seconds become hundreds.
Based on other threads, โrun-time information collectionโ should be disabled to improve performance. Well .. it already is .. so what else can you do (if anything)?

source
share