Example of changing numba configuration for RDST

This example shows how to modify the numba options to change the behaviour of numba functions

Load the module and change the numba options

The __init__.py file of convst contains four variable that will change how numba functions are compiled, please refer to the numba documentation for the effect of each option. In a classic desktop use of convst, you should not need to change any option. Issue have been known to occur on some HPC cluster. By default, the variables are all defined as True, here we change the value of __USE_NUMBA_CACHE__ and __USE_NUMBA_PARALLEL__ to False.

import convst

convst.__USE_NUMBA_CACHE__= False
convst.__USE_NUMBA_FASTMATH__ = True
convst.__USE_NUMBA_NOGIL__ = True
convst.__USE_NUMBA_PARALLEL__ = False

Run convst with the modified numba options

We can now use convst with these options, to check if the changes have been taken into account, we can inspect a numba function argument. !! THE MODIFICATION OF THE NUMBA CONFIGURATION MUST BE MADE BEFORE CALLING ANY NUMBA FUNCTION, NUMBA TREAT GLOBAL VARIABLE AT CONSTANT AT COMPILE TIME CHANGING THE VALUE OF THESE PARAMETER AFTER A FUNCTION IS COMPILED WILL NOT CHANGE ITS BEHAVIOR.

from convst.transformers._univariate_same_length import U_SL_apply_all_shapelets
print(U_SL_apply_all_shapelets.targetoptions)