FF optimization in conjunction with CSP

[qzhu2017]

Example 1:

Use aspirin as an example:

1. python 1-on-the-fly-csp.py -g 100 -p 128 -n 128 --ffopt > log1-ff1-opt
2. python 2-opt-ff2.py > log2-ff2-opt
3. python 3-csp-ff1-ff2.py -g 200 -p 128 -n 128 > log3-csp
4. python 4-test-ffs.py > log4-test

[qzhu2017]

Change on-the-fly mode to save mode.
i.e., no update of the ff parameters,
only choose to save references

1. run ff-csp for 10 generations, at the end of each generation, randomly pick a few structures, make augmentation (1->16), accumulates the references.xml

2. update ff with PSO-optimization, save pso.xml

3. run ff-csp for 10 generations again, at the end of each generation, [some strategy] to pick a few structures, make augmentation (1->16), continue to accumulate the references.xml

4. load, pso.xml from the previous run, update ff, save pso.xml

5. repeat 3-4 for a few iterations.

Goals

1. The priority is to make it run faster with parallelization
2. [some strategy]: have a some balance between low-energy and high uncertainty.

[qzhu2017]

To do

• run save PSO from the previous results

[qzhu2017]

• Remove high energy structure wrt to ref_energy
• Move cpu-*** folder to the subfolder