SPR difference stalls

[steven.wad...@gmail.com]

Hi All,

Has anyone ever experienced the sprdiff command to cause TNT to go "not responding"? I find it happens on rare occasion... maybe it's sensitive to the complexity of difference between trees? 

I've found that this hang up can be sensitive to number of replications... for instance if I compare the two trees that are giving me problems today, and use the default N=10, TNT never finishes the process. However, if I reduce the reps to 4, it finishes quickly.

Presumably more reps is more accurate. I get a lower % similarity using 2 reps than using 4. If I use 5, TNT gets hung up.

Has anyone had this issue and figured a way to overcome it?

Many thanks!

Steven

[steven.wad...@gmail.com]

Hi,

With some more trials, I found that I can get TNT hung up by simply increasing the reps... even if comparing trees that I previously thought there were no issues.

If I use:

sprdiff 0 1 10; (finishes less than 1 sec, reports 21 moves)

sprdiff 0 1 50; (finishes less than 1 sec, reports 20 moves)

sprdiff 0 1 60; (finishes less than 1 sec, reports 20 moves)

sprdiff 0 1 65; (never finishes)

Clearly it's better to increase the reps to more than the default of 10, because there seems to be swapping schemes that find less moves to transform one tree into the other. But it also seems that something about the sprdiff command gets bogged down if the reps are too high under a certain comparison.

I've also found that different tree pairs have a different rep threshold where TNT gets hung up.

I'd really like to be able to set reps to 500 or 1000 to have a little more confidence that I'm catching the fewest possible swaps (highest similarity estimate). 

Any thoughts?

Cheers,

Steven