Fix occasional error in debias#161
Conversation
| self.ICv = DICTIONARY['IC']['vox'] | ||
| self.ICo = DICTIONARY['IC']['dir'] | ||
| self.ICeval = DICTIONARY['IC']['eval'] | ||
| self.ICm = DICTIONARY['IC']['mask'] |
There was a problem hiding this comment.
What about DICTIONARY['IC']['debias_mask]?
| unsigned int [::1] ICv | ||
| unsigned int [::1] ECv | ||
| unsigned int [::1] ISOv | ||
| unsigned int [::1] ICeval |
There was a problem hiding this comment.
I would avoid the underscore and keep it shorter, since all the others are in the form "compartment + one letter" (e.g., ICv, ICo, ICf, ...)
|
|
||
|
|
||
| def build_operator( self ) : | ||
| def build_operator( self, mask_ic=None ) : |
There was a problem hiding this comment.
What about ic_debias_mask? In the future, we might also have ec_debias_mask etc
| debias : boolean | ||
| If true, a debiasing step will be performed after the main fitting | ||
| procedure. Highly suggested when using a regularisation. (default : False) | ||
| debias_cond : float |
There was a problem hiding this comment.
This is de facto a threshold; what about debias_thr?
| logger.info( 'Running debias' ) | ||
| logger.subinfo('') | ||
| logger.info( f'Running debias' ) | ||
| logger.subinfo( f'Creating mask for IC compartment with the condition: <={debias_cond:.2e}', indent_lvl=1, indent_char='*' ) |
There was a problem hiding this comment.
Perhaps, to be clearer, we can write Keeping columns whose corresponding x.....?
There was a problem hiding this comment.
Is Keeping columns in the IC compartment whose corresponding coefficient is greater than {debias_thr} working well for you?
2 participants
Fix error in saving results (RMSE and NRMSE maps) after debias, happening with some datasets when using BS or SZB models (Fixes #159).