Speed up quad sus tf calculation 10x, by pre-solving the system symbolically

Profiling of precompIFO() before this change:

%timeit gwinc.precompIFO(freq, ifo)

404 ms ± 24 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%prun ifo2 = gwinc.precompIFO(freq, ifo)

      693645 function calls (693211 primitive calls) in 0.833 seconds

Ordered by: internal time

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
 30000    0.362    0.000    0.695    0.000 linalg.py:319(solve)
    10    0.083    0.008    0.778    0.078 suspension.py:39(calc_transfer_functions)
 30008    0.065    0.000    0.109    0.000 linalg.py:141(_commonType)
 30018    0.056    0.000    0.056    0.000 {method 'astype' of 'numpy.ndarray' objects}
 60012    0.046    0.000    0.098    0.000 linalg.py:108(_makearray)
 etc, etc

And after:

%timeit gwinc.precompIFO(freq, ifo)

28.6 ms ± 4.87 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

%prun ifo2 = gwinc.precompIFO(freq, ifo)

      3596 function calls (3162 primitive calls) in 0.032 seconds

Ordered by: internal time

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    74    0.016    0.000    0.016    0.000 coatingthermal.py:743(getCoatRefl)
    10    0.006    0.001    0.006    0.001 suspension.py:30(tst_force_to_tst_displ)
     1    0.005    0.005    0.012    0.012 suspension.py:51(suspQuad)
   137    0.001    0.000    0.001    0.000 {built-in method numpy.core.multiarray.zeros}
     2    0.001    0.000    0.001    0.000 suspension.py:40(top_displ_to_tst_displ)
etc, etc

By the way, the next biggest hog is the TF products in quantum.py:

%timeit gwinc.noise_calc(ifo2, freq)

      40613 function calls in 0.181 seconds

Ordered by: internal time

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
 21000    0.074    0.000    0.074    0.000 {built-in method numpy.core.multiarray.dot}
     7    0.056    0.008    0.137    0.020 quantum.py:409(getProdTF)
     1    0.020    0.020    0.088    0.088 quantum.py:257(shotradSignalRecycled)
     2    0.007    0.003    0.007    0.003 coatingthermal.py:446(getCoatTOPhase)
     6    0.004    0.001    0.005    0.001 coatingthermal.py:234(getCoatThickCorr)
  9003    0.003    0.000    0.003    0.000 {method 'squeeze' of 'numpy.ndarray' objects}

Looks like the quantum code can speed up about 5x if we replace the looped np.dot with np.matmul. Let me clean that up a bit and I'll make a separate merge request for it.

mentioned in merge request !32 (merged)

Can you add whatever notebook you used or a pdf/html rendering of it into the doc directory?

It seems if we are going to use generated code/math we should have the ability to regenerate it, as it is otherwise much more opaque. I don't think it needs to be particularly clean or documented, and it can be some work for someone to attempt to run it again, but "the code behind the code" seems like a good thing to include.

Again, I'll try to check this out and merge soon.

added 1 commit

• 4555bfc8 - Include sus TF generating code -- now we're ready for the quintuple pendulum

Compare with previous version

added 4 commits

• 4555bfc8...ea42d0c6 - 2 commits from branch master
• cc89719d - Speed up quad sus tf calculation 10x, by pre-solving the system symbolically
• 442ac0b7 - Include sus TF generating code -- now we're ready for the quintuple pendulum

Compare with previous version

merged

mentioned in commit e9aabda9