README

We document 2 programs: refinedate and gsmooth 

1) 
refinedate -a <rawadate> [-x <refadate>] -b <rqawbdate> -y <refbdate>  -d <disfile> -z <fulljoint> 
rawadate and rawbdate are initial prior estimates.  
Format 2 column data: 

date probability
The second column can be a histogram (count) as refinedate normalizes to sum to 1.  

Convention:  more ancient dates are more negative.  Use BP. or BCE. NOT CE


disfile  
Distribution of difference of date.  
Prior distribution of (death date of A) - (death date of B) 
Note:  If we expect that A dies more recently this will be positive (in the mean) 
fornmat: 
<date difference>  probability 
Again, a histogram (count) is acceptable in the last column.  

It is up to the user to supply this table
but 4 suitable tables are provided in .../reldates 

*** all 3 input files need to be at a resolution of 1 year *** 


================
gsmooth -i <raw> -o <smooth> [-s] [-f filterlength]  

Smooths the raw file which can be coursely quantized (say at 5 year intervals).  
smooth is the output file 

Optional parameters: 

-f filterlength :: apply a rectaangular filter,  filterlength must be odd.
So for eeample if filterlength is 3 we average 3 conseitive values of <raw> and replace the central value.  
This damps down spurious "wiggles" but does tend to slightly reduce the peak.  

-s   force symmetry.   For some relative pairs, such as siblings, there may be no information about birth order, 
and so it is appropriate to force the distribution of disfile to be symmetric.  

../reltables/mksmooth shows how our tables were made and the parameters used for gsmooth

More information is in ./refinedate.pdf 

Nick  Patterson  April 2021