tflovorn/scexplorer

Exploration of the parameter space of a new model of cuprate superconductors

scexplorer's Issues

specific heat plots - too many x's

After calculating specific heat, plotting finds X values sometimes split into several nearby values.

It would be nice to fix this in the plot package, but that seems tough (it uses floats as map keys in ExtractSeries()).

Fix this by holding on to original X values, then iterating through envs after SH calculation. Set each X to its closest neighbor in the original X value list.

performance enhancements suggested by profiling

Profile with:

go test -c
go test -test.cpuprofile cpu.out
go tool pprof packageName.test cpu.out

in pprof:
top10, top40, etc. get most common functions
top40 -cum to sort by cumulative (+children) time
web to visualize

---

For tempFluc -testPlot, got:

31.7% of time spent in malloc through ZeroVector calls resulting from Vector.Add and Vector.Mul

14.3% of time spent in math.Tanh
12.9% of time spent in math.Sin

Sin calls are Sin(kx), Sin(ky); could cache these results. Vector Add/Mul allocation could be avoided by requiring result vector to already be allocated; most of the time it will be.

plots design

It would be nice to have a matplotlib interface that gives us an image. Pass a channel to the plotting function; when the plot is done, it returns the plot along with any errors via the channel. (the function could also just return this channel)

multiplot should(?) run with one python process

It seems very inefficient to spawn lots of Python processes to handle MultiPlot. Is there a way to measure the performance impact of this?

benchmark cached package

It would be nice to know how much faster cached.Cos and cached.Sin are for typical use case (within bzone reduction with constant pointsPerSide and dimension) compared to math.Cos and math.Sin.