Resuming iteration

Context: Common datafile line qualifiers

!CONTINUE

!CONTINUE [rsvfile] is used to restart/resume iterations from the point reached in a previous run. This qualifier was originally defined for the datafile line but is now usually set from the command line using the option letters C (continue) or F (final) options, or the top job control line using !CONTINUE or !FINAL.

!FINAL on the top control line causes ASReml to do one iteration starting from the values in the current .rsv file. You would do this when the only change to the job was the adding of new predict statements.

After each iteration, ASReml writes the current values of the variance parameters to a file with extension .rsv (re-start values) with information to identify individual variance parameters. The !CONTINUE qualifier causes ASReml to scan the .rsv file for parameter values related to the current model replacing the values obtained from the .as file before iteration resumes. If the model has changed, ASReml will pick up the values it recognises as being for the same terms. Furthermore, ASReml will use estimates in the .rsv file for certain models to provide starting values for certain more general models, inserting reasonable defaults where necessary. The transitions recognised are
  • DIAG to FA1
  • DIAG to CORUH (uniform heterogeneous)
  • CORUH to FA1 or to XFA1
  • FA i to FA i+1
  • XFA i to XFA i+1
  • FA i to CORGH (full heterogeneous)
  • XFA i to US (full heterogeneous)
  • CORGH (heterogeneous) to US When running a job with multple parts (see !DOPART with !RENAME and !ARGS ), the part number is built into the file name. If the required .rsv file does not exist, ASReml will use the .rsv file from the most recent part run.

    !CONTINUE file

    !CONTINUE file is used to restart/resume iterations from the parameter values in the named non-current .rsv file. This is intended for use with !CYCLE to cause ASReml to start each cycle using the values in the file, rather than the values written to the current .rsv file from the last cycle.

    !TSV and !MSV

    !TSV !MSV is used to restart/resume iterations from the parameter values obtained from the current .tsv or .msv file. run. ASReml 4 includes a modified form of the .rsv file which is accessible to the user. The .tsv file is written with the initial values ASReml will use in the first iteration. Use !MAXITER 0 to create the .tsv file without performing any iterations. The values in the .rsv file may have been supplied by the user in the .as file, determined by ASReml as default values or retrieved from an .rsv file. The .msv file has the same format as the .tsv file, but contains the parameter values obtained in the final iteration. These files may be edited, taking care not to change the format, but changing parameter values, and used for input. Use !TSV on the top control line to take start values from the .tsv file, !MSV on the top control line to take start values from the .msv file. As an example 
     Wolfinger Rat data
 treat !A
 wt0 wt1 wt2 wt3 wt4
 subject * !=V0
 wolfrat.dat !skip 1 !ASUV !MAXITER 0
 wt0 wt1 wt2 wt3 wt4 Trait treat Trait.treat
 residual units.us(Trait)


     # This .tsv file is a mechanism for resetting initial parameter values
 # by changing the values here and rerunning the job with !TSV
 # You may only change values in the last 4 fields.
 # Fields are:
 # GN, Term, Type, PSpace, Initial value, RP GN, RP scale.
 5, "units.us(Trait);us(Trait) 1", G, P, 4.7911110 , 5, 1
 6, "units.us(Trait);us(Trait) 2", G, P, 5.0231481 , 6, 1
 7, "units.us(Trait);us(Trait) 3", G, P, 15.298889 , 7, 1
 8, "units.us(Trait);us(Trait) 4", G, P, 4.8438271 , 8, 1
 9, "units.us(Trait);us(Trait) 5", G, P, 11.264815 , 9, 1
 10, "units.us(Trait);us(Trait) 6", G, P, 26.095692 , 10, 1
 11, "units.us(Trait);us(Trait) 7", G, P, 4.6882715 , 11, 1
 12, "units.us(Trait);us(Trait) 8", G, P, 10.824074 , 12, 1
 13, "units.us(Trait);us(Trait) 9", G, P, 27.332887 , 13, 1
 14, "units.us(Trait);us(Trait) 10", G, P, 71.875403 , 14, 1
 15, "units.us(Trait);us(Trait) 11", G, P, 3.9083333 , 15, 1
 16, "units.us(Trait);us(Trait) 12", G, P, 10.292592 , 16, 1
 17, "units.us(Trait);us(Trait) 13", G, P, 34.137962 , 17, 1
 18, "units.us(Trait);us(Trait) 14", G, P, 69.287036 , 18, 1
 19, "units.us(Trait);us(Trait) 15", G, P, 141.97296 , 19, 1
    Parameter constraints and initial values can be changed by editing the values in the PSpace and Initial value columns. Scale relationships can be introduced by noting that the relationships can be written as parameter = subset parameter * scale or GN column, RP GN column * RP scale column. The relationships generated by !VCC 2 
     5 6 8 11 15 7 * 2 9 * 2 12 *2 16 *2
 #parameters 7 9 12 16 equal and  twice 5 6 8 11 15
 10 13 17 #parameters 10 13 17 equal
    can be introduced by editing the RN GN and RP scale columns so that the data values in the .tsv file become 
    <5,>"units.us(Trait);us(Trait) 1", G, P, 1.00000000 , 5, 1.0000
<6,>"units.us(Trait);us(Trait) 2", G, P, 1.00000000 , 5, 1.0000
<7,>"units.us(Trait);us(Trait) 3", G, P, 2.0000000 , 5, 2.0000
<8,>"units.us(Trait);us(Trait) 4", G, P, 1.00000000 , 5, 1.0000
<9,>"units.us(Trait);us(Trait) 5", G, P, 2.0000000 , 5, 2.0000
<10,>"units.us(Trait);us(Trait) 6", G, P, 1.00000000 ,10, 1.0000
<11,>"units.us(Trait);us(Trait) 7", G, P, 1.00000000 , 5, 1.0000
<12,>"units.us(Trait);us(Trait) 8", G, P, 2.0000000 , 5, 2.0000
<13,>"units.us(Trait);us(Trait) 9", G, P, 1.00000000 , 10, 1.0000
<14,>"units.us(Trait);us(Trait) 10", G, P, 1.00000000 , 14, 1.0000
<15,>"units.us(Trait);us(Trait) 11", G, P, 1.00000000 , 5, 1.0000
<16,>"units.us(Trait);us(Trait) 12", G, P, 2.0000000 , 5, 2.0000
<17,>"units.us(Trait);us(Trait) 13", G, P, 1.00000000 , 10, 1.0000
<18,>"units.us(Trait);us(Trait) 14", G, P, 0.20000000 , 18, 1.0000
<19,>"units.us(Trait);us(Trait) 15", G, P, 1.00000000 , 19, 1.0000

    If !TSV (or !MSV) is specified ASReml will read the current (created with the same PART number) .tsv (or .msv) file. If there is no current .tsv (or .msv file), a non-current