H I S T O R Y O F M I G R A T E
[people] in brackets helped with resources or reported problems.
------------------------------------------------------------------------
2001
-------------------------------------------------------------------------
August 15 Migrate-n 1.2.4 fix for geographic distance file option, some older versions [most likely
version that were newer than 1.1] were using
a similarity matrix instead of a distance matrix.
August 9 Migrate-n 1.2.3 On Compaq alpha the program crashed with
underflows (exp(-1000)=NaN and not 0 as everywhere else),
now every exp() will be safeguarded on Compaq alphas.
August 8 Migrate 1.2.2 Several minor fixes to "beautify" outfile and
menu printing.
July 28 Migrate-n 1.2.1 changes to the menu to incorporate AIC for
migration model selection, needs documentation.
July 15 Migrate-n 1.2 Reworking of microsat likelihood calculation.
It seems that my change on April 15 was only doing half
of the job, the conditional likelihood calculation
were not taking into account unobserved alleles above and
below the smallest and largest repeatnumber.
Everybody using microsatellite data should upgrade.
April 30 Migrate-n 1.1 Reworking of likelihood calculation,
should speed up parameter estimation about 10-20%.
Inconsistency between manual likelihood-ratio description
and program fixed, manual is now at version 1.1.
If you want to use likelihood ratio test you should
run 'long' runs or then use replication or heating.
April 20 Migrate-n 1.0.4 Harmonizing the use of end-of-line
characters, migrate had several problems
reading files that were moved between different
operating systems, hope this is all fixed now.
Fix of FAST and QUICK profile options [Raphaelle Chaix].
April 18 Migrate-N 1.0.3. More fixes, plots should work now
[Steven Irvin], windows binary now is compiled with
correct set of C-files and not old ones [Mats Bjo"rklund].
April 15 Migrate-N 1.0.2 Serious problem in microsatellite code
found and fixed: the stepwise-mutation model was only
approximated due to a programming error. The probability
of the number if steps for a given length of time was only
too crudely approximated. Without the error reports of
Steven Irvin and Raphaelle Chaix I would not have found
this. Problem with "?" as the last character in a file
transported between different operating systems fixed
[R. Chaix]. Error in new replicate-summarizing code fixed
[S. Irvin].
April 11 Migrate-N 1.0.1 Glitch in single locus likelihood ratio test
fixed (technically you should not use this anyway).
Some small formatting issues when saving parmfile.
April 10 MIGRATE-N 1.0 several memory leaks fixed, plotting
option changed and will plot only over all loci.
Many internal changes/speedups that should not affect users,
although may introduce no bugs. Gamma deviated mutation rate
among loci changed considerably but still has difficulties
to converge. Sumfile run with "allelic" data broke,
fixed now.
February 12 MIGRATE-N 0.9.14 some cleanup on freeing memory in profile
evaluation. Fixed a bug in the heating code when the dataset
is "allelic data" [Robb Brumfield].
------------------------------------------------------------------------
2000
-------------------------------------------------------------------------
December 16 MIGRATE-N 0.9.13 Speed up: memcpy() in site rate category code
is replaced by pointer swapping [C++ newsgroup contribution],
more precalculation of parts of the acceptance-ratio
calculation. Can use now a geographical distance file to
specify a kind of an isolation by distance model.
Rearrangement of likelihood ratio test menu [Steven Irvine],
removal of the l-ratio=LOCUS option, l-ratio=MEAN is replace
by l-ratio=MLE, this actually should not break old parmfiles.
December 05 MIGRATE-N 0.9.12 Bug fix: I broke the infinite allele code
and it is fixed now again [Ron Goldthwaite]
November 26 MIGRATE-N 0.9.11 Bug fix: when compiled optimized on SUNs
the program showed odd and wrong behavior, and did not
accept any new genealogy after a while, perhaps it is
a compiler problem, because it does not occur on Linux
Intel and I have not found a memory problem. A more
restrictive array copying seems to remedy the problem.
Additions: Adapted for compiles in a MAC OS X terminal
window, it runs about twice as fast as in MAC OS 9.
Addition of an option for more accurate calculations
during the data likelihood calculations with large data
sets where the individual probabilities underflow and the
program crashes, this option is not necesary for many
data sets, but slows the down run around 20-40%.
October 18 MIGRATE-N 0.9.10 Bug fix: Multiple microsatellite loci analyzed
with Brownian motion model, populations with no data, and a
missing value in at least one of the sampled populations
crashed, because the whole locus was discarded, it should
work properly now [Martin Damus]
October 13 MIGRATE-N 0.9.9 Bug fix: Parallele evaluation on symmetric
multiprocessor machines now works with rate categories (more
than one process wrote to the same unguarded memory).
Further redesign to easy transition to parallele processing
of loci.
October 6 MIGRATE-N 0.9.8 Bug fix: The reanalysis of a sumfile
with one locus failed in the profile likelihood calculation.
Addition of saveguards for machines (Dec Alphas)
that return for exp(-1000) = NaN instead of a very small
number or zero.
Addition of a sequencing error
possibility [see under Datatype with Sequence data].
Heating scheme expanded from 4 chains to n<20 chains.
July 28 MIGRATE-N 0.9.7 Bug fix: in cases where a population size
cannot be well estimated (the likelihood surface is flat)
the reset function failed to calculate an average size,
and returned 0.0 which resulted in erratic behavior
[Patricia Brito].
July 22 MIGRATE-N 0.9.6 Addition of a logfile option,
the Gamma-deviated mutation rate among loci seems to work
but needs more rigorous testing, so sometimes it will still
fail.
July 11 MIGRATE-N 0.9.5 Bug fixes: the addition of a null population
should work now for all datatypes [Martin Damus],
under some conditions the maximizer
gave up too quickly, and (an embarassing one) for profile
likelihood percentiles miscalculation of percentile values:
some of the old percentiles were wrong, To see what impact it
had on your conclusions see below
correct/:1% 5% 10% 25% 50% 75% 90% 95% 99%
wrong/old: 0.5% 2.5% 5% 12.5% 50% 87.5% 95% 97.5% 99.5%
The old tables were using the 1,5,10... labels but calculated
values under "wrong/old".
[the likelihood ratio tests are not affected by this]
The new profile tables are set so that you can generate
99%, 95%, 90%, 50% confidence intervals.
[mutation=Gamma is still broken, sigh]
May 30 MIGRATE-N 0.9.4
Fixed a bug in reading and writing summary files
(options affected were write-summary and datatype=genealogy).
mutation=Gamma is still broken [Eric Simandle],
do not use it.
May 12 MIGRATE-N 0.9.3
embarrased to say but the last fixed introduced a problem,
in the likelihood calculation, hopefully fixed now.
mutation=Gamma is still broken [Eric Simandle],
do not use it.
April 22 MIGRATE-N 0.9.2
inconsistency in likelihood calculation with replication
fixed.
April 21 MIGRATE-N 0.9.1
Bug in Mac-version of automatic random number seed
generation, and in recording start migration parameters fixed,
and migration start parameter mix up in parmfile
fixed [all Ken Wahrheit].
Heating scheme changed, implemented a 4 parallel chain
heating scheme (simulated tempering) based on Geyer and
Thompson. The Tempered transition method (Neal) will
be reimplemented in a later version.
Fixes: ttratio now works for different values
[Judite Alves],
Registered users: 423
(tried to find this time all doubles)
March 3 MIGRATE-N 0.9
First introduction of estimation of parameters
over multiple chains or multiple runs.
Problems: Multiple chain/runs
with the combination of gamma deviated mutation rate
does not work yet. Heating scheme is broken.
------------------------------------------------------------------------
1999
-------------------------------------------------------------------------
December 10 MIGRATE-N 0.8.5
Change of defaults: plot=FALSE, moved eventloop()
in plot routine for Macintosh.
December 2 MIGRATE-N 0.8.4
Revision of likelihood ratio test output. Change
of "burn-in" default from 200 to 10000.
Minor speedups in several functions.
November 23 MIGRATE-N 0.8.3
Revision of heating scheme. But still needs more testing.
November 5 MIGRATE-N 0.8.2
Addition of a convergence criterium: Gelman's R,
(use progress=verbose)
Added material to the
likelihood ratio test documentation.
Several minor bugfixes (sumfile related [Tonya Bitner],
Profile Quantile table, verbose Progress reporting)
Registered users: 372
September 7 MIGRATE-N 0.8.1
More cleanup of C-code, incorporation of new spline
routine ( but this is still experimental). Improvement
of documentation.
August 20 MIGRATE-N 0.8
A problem with the UPGMA starting tree fixed,
with many individuals the starting tree contained
some silly ordering, that produced uneven number of
migration events on this tree and needs rather a long
time to recover from this.
profile likelihood speed improvements when there is a
custom-migration matrix with zeroes.
Registered users: 322
June 4 MIGRATE-N 0.7.1
Division by 0 bug fixed in fst-calculation, this seems
to bother only DEC Alphas.
June 1 MIGRATE-N 0.7
Updated documentation, several minor things, warnings and error
reporting should be more consistent, I am adding a section to
the manual that describes all error/warning messages [partly
done], the plotting graphics are more flexible now, but still
need more work. You can specify the range and type of
axes (log-scale, std-scale), and if the migration parameter
shall be plotted as M=m/mu or 4Nm. Fix of inconsistency
in migration value menu input [Reinaldo Brito].
Fix of an error in the
profile-method=FAST (it will need now more time to finish,
because it is doing the final maximization over all other
parameters), if you want its old behavior, that assumes that
Theta and M are not correlated [not a too bad assumption],
then use profile=YES:QUICK.
March 8 MIGRATE-N 0.6.3
Updated documentation (fixed errors in description of
random-seed options, added important material
to profile-likelihood) ,
inclusion of improved man page,
fixed configure for SGI's with out gcc.
Feb 14 MIGRATE-N 0.6.2
Tree traversal debug code removed,
this killed runs with many individuals [Lisle Gibbs]
Configure for SGI changed, does it work?
MIGRATE-0.4: no change.
Registered users: 280.
------------------------------------------------------------------------
1998
-------------------------------------------------------------------------
December 29 MIGRATE-N 0.6.1 Multilocus estimates are all wrong
in version 0.6, silly programming mistake found an fixed.
If you have used microsatellites or electrophoretic markers
or several sequence loci you need to rerun that analysis.
Result table should print now nicer with population numbers
above 3.
MIGRATE-0.4: no change.
Registered users: 234
Oct 29 MIGRATE-N 0.6
Addition of datatype=n that is for single nucleotide
polymorphism data, no simulation with this kind of data
is yet done, so I do not know about biases etc.
Profile tables now report 4Nm instead of m/mu for
the migration parameters.
Documentation changed and contains now more about
how to read the outfile and what you can and
cannot do with the reported log(likelihood) values
[Mats Bjorklund].
Binaries for OPENSTEP available [thanks to Magnus Nordborg
giving me an account on his machine].
Registered users: 206
Sep 1 MIGRATE-N 0.4/0.5 [was not released, was too busy with
other things]
FST start values work now also for microsatellite data
but I still need to check the correctedness of the FST table
when the data are microsatellites.
Fixed wrong emmigration plots. Fixed wrong start
calculations for allelic data when a delimiter was used,
and several minor bug fixes. Profile-method
"uncorrelated" from version alpha.1 recovered.
Registered users: 197
June 14 MIGRATE-N alpha.3 and MIGRATE-0.4.2
Several minor changes in migrate-n: menu addition for
-profile method:
profile-method=<Spline | Percentiles | Discrete>
Spline: uses 1-dimensional splines to find percentiles,
faster than the "Percentiles" option but not so accurate,
"Discrete" evaluates at "fixed" (0.02, 0.1, 0.2, 0.5, 1, 2,
5, 10, 50) * MLE of parameter.
-with progress=yes you can see now a rough prognosed time
of end of sampling genealogies and if you use profiles
an estimated time of finishing.
-Fix of reading in intermediate results (sumfile).
-Most importantly a (hopefully) stable compile for
Windows, I failed to find the cause why the program
compiled with WATCOM failed to finish with "bigger" data sets,
it is now compiled with mingw32/gcc-win32, this is
a windows port of the same system I am using on my workstation.
Please report failures, I can only try a limited set of
examples.
Migrate-0.4.2: new windows binary (using mingw32/gcc-win32)
Registered users: 163
May 30 MIGRATE-N alpha.2 and MIGRATE-0.4.1
With more than 2 sequence loci, there was a problem
with the T/T-ratio, when the ratio was not specified
for each locus.
Start parameter problems with microsatellite data fixed
[Mats Bjorklund].
Persistent problems with Windows executable
sometimes I get floating point errors, on all other systems
this does not occur.
Registered users: 153
May 29 MIGRATE-N alpha.1 and MIGRATE-0.4
Memory bug in FST calculation found and fixed
[Daniel Yeh]
No change of Migrate-0.4
Registered users: 148.
May 26 MIGRATE-N and MIGRATE-0.4
This release has the two population version (Migrate-0.4)
and an alpha-version of Migrate-n that can solve migration
matrix population model with unequal population sizes and
unequal migration rates for n populations, I tried up to 10
and the results where fine, but I am pretty sure that if
you try to feed in all your date of 100 subpopulation it
will (a) probably crash, but more importantly (b) will
need TERRIBLY long to run.
I would like to get some
feedback about what you want to see in the outfile,
menu etc. Registered Users: ~138.
February 25 MIGRATE 0.4 (was not put up onto the website, I was to busy)
More complex sequence evolution models (categories,
weights, autocorrelation etc.) should work now,
it was broken. Cleanup of some output file lines, and
some menu entries. The FST estimation (Remember FST is only
used to generate start parameter values) is
in pre 0.4 versions logically flawed. It estimates
2 parameters per population using F_within and F_between,
but there is only 1 F_between. Correctly, we can
only estimate maximally 3 parameters with 1 locus for
two populations. I added an option into the MENU and into
the PARMFILE (fst-type=<Theta | Migration >) with which you
can decide which parameter is considered the same for both
populations.
Registered users:89
------------------------------------------------------------------------
1997
------------------------------------------------------------------------
August 20 MIGRATE 0.3.1
Confusing menu entries for start theta and 4Nm values
fixed [Carol Reeb], the start migration values are now
4Nm and *not* m/mu values as before. Automatic Random
number seed on Macs and perhaps on other Systems
delivered sometimes negative values, now fixed
[Carol Reeb], although I would recommend to use your
own random number seeds: best values are 4n + 1 in the
range of 5 .. 2147483647, so there are plenty of
start random number seeds. Menu entry for
usertree options should be now more clear, the usertree
options needs a genealogy with migration events on
it [Tony Metcalf]. Currently MIGRATE can construct
those, or you have to do it by hand, if you need to do
this send me email, because the doc is not updated.
Registered users:52
June 20 MIGRATE 0.3.0.
Brownian motion approximation to stepwise mutation
model for microsatellites added. Solved problems:
Input problems with microsatellites data, major memory
allocation problem for datasets with more than 100
gene copies fixed [Carol Reeb]. Update of some
citation and FST output tables [Byron Adams].
Persistent problems: Long sequences AND high number of
individuals need nuch longer chains than the proposed
default. Try ten times longer "long" chains. Or use
the option "moving-steps".
Registered users:38
May 12 MIGRATE 0.2.1a.
Fixed problems: Interleaved sequence data
should work now, last character of individual names is
now printing, and printing of second population data
should work, too, although the EP data printout is
still ugly. [Allen Rodrigo]. Memory problem with some
Allelic data fixed.
Registered users: 30
April 30 MIGRATE 0.2a released.
Fixed problems or changes: Corrections of several
minor problems, Printing of the data fixed, but still
ugly; Memory problem with large sequences fixed.
Options: treefile added, can write now a genealogy
with migrations; the option progress=Verbose for more
information during a run, the progress=Yes gives now
less information than before. Output: covariance
matrix for combined loci now prints, too. Persistent
problems: -Long sequences need very long chains to
remove the starting conditions for the migration rate
from the first tree (see documentation).
-Microsatellites still have probably a bias downwards
in Theta, but I need more simulations to make this
more clear.
Registered users: 8
March 4 First trial release of MIGRATE 0.1a
This release is not announced widely, because I have
to test, almost everything including all HTMLs,
registration, and the program itself: simulations need
time. Registered users: 1
--------
$Id: HISTORY,v 1.39 2001/05/17 23:55:29 beerli Exp $