annealing

Tries to optimize the maximum multipoint loglikelihood using a dedicated simulated annealing stochastic optimization algorithm.

Synopsis:

The annealing command is invoked either as:

• annealing Options
• annealing LPlateau InitTemp FinalTemp Cooling

Description:

The annealing command tries to improve the loglikelihood of the best known order for the current list of active loci (the best map in the heap) using a dedicated simulated annealing search algorithm. The annealing simulation starts from the temperature InitTemp and repeats until the temperature FinalTemp has been reached using a geometric cooling schedule with a cooling parameter Cooling and constant temperature plateaus of length LPlateau. The initial temperature is automatically reajusted if it is too cold/hot.

During the search, the algorithm gives some feedback by printing the current temperature and a character ``+'' if an imporved map has been found.

Arguments :

• Options : -u to obtain the synopsys of the normal use, -h to print a one line description, -H to print a short help.
• LPlateau: length of constant temperature plateaus in the colling schedule.
• InitTemp FinalTemp: starting/ending temperature for the simulated annealing.
• Colling: cooling schedule parameter. This parameter should be between $0$ and $1$ and is typically taken to be close to $1$. The closest to $1$ the longer but also the better the algorithm.

Returns:

nothing. All the map explored by the algorithm are candidate for the heap.

Example:

# we first load a dataset
CG> dsload Data/rh.cg
{1 haploid RH 53 118 /homes/thomas/carthagene/test/Data/rh.cg}
# perform linkage group detection (output omitted)
CG> group 0.3 3
...
# select group number 10
CG> mrkselset [groupget 10]

# put a (stupid) map in the heap
CG> sem

Map -1 : log10-likelihood =  -161.87
-------:
 Set : Marker List ...
   1 : G5 G18 G17 G14 G16 G13 G12 G6 G7

# use a (fast) annealing command
CG> annealing 100 100 0.1 0.9

Map -1 : log10-likelihood =  -161.87
-------:
 Set : Marker List ...
   1 : G5 G18 G17 G14 G16 G13 G12 G6 G7

100.00? :  +++
90.00? :
81.00? :  +
72.90? :
65.61? :
59.05? :  +
53.14? :
47.83? :
43.05? :
38.74? :
34.87? :
31.38? :
28.24? :
25.42? :
22.88? :
20.59? :
18.53? :
16.68? :
15.01? :
13.51? :
12.16? :
10.94? :
9.85? :
8.86? :
7.98? :
7.18? :
6.46? :
5.81? :
5.23? :
4.71? :
4.24? :
3.82? :
3.43? :
3.09? :
2.78? :  +
2.50? :
2.25? :
2.03? :
1.82? :
1.64? :
1.48? :
1.33? :
1.20? :
1.08? :
0.97? :
0.87? :
0.79? :
0.71? :
0.64? :
0.57? :
0.52? :
0.46? :


# we check the map with a large flip
CG> flips 9 0 0

Single Flip(window size : 9, threshold : 0.00).


Map -1 : log10-likelihood =  -141.81
-------:
 Set : Marker List ...
   1 : G7 G12 G13 G16 G6 G5 G17 G18 G14


 2 2 2 3 2 1 3 3 2  log10
 1 6 7 0 0 9 1 2 8   -141.81
[8 7 6 5 - 3 2 1 0]      0.00


# the map found was indeed optimal

See also:

• greedy (2.6.8)
• flips (2.6.4)
• polish (2.6.5)