def _remote_bin_iter(iiter, n_iter, dsspec, wt_dsspec, initpoint, binbounds,
ignore_out_of_range):
iter_hist_shape = tuple(len(bounds) - 1 for bounds in binbounds)
iter_hist = numpy.zeros(iter_hist_shape, dtype=numpy.float64)
dset = dsspec.get_iter_data(n_iter)
if dset is None:
return iiter, n_iter, iter_hist
else:
npts = dset.shape[1]
weights = wt_dsspec.get_iter_data(n_iter)[:, 0, 0]
#dset = dset[:,initpoint:,:]
for ipt in range(npts - initpoint):
histnd(dset[:, ipt, :],
binbounds,
weights,
out=iter_hist,
binbound_check=False,
ignore_out_of_range=ignore_out_of_range)
del weights, dset
# normalize histogram
normhistnd(iter_hist, binbounds)
return iiter, n_iter, iter_hist
def _test_uint(npts=1024 * 1024, ndim=3, loops=3):
from time import time
mine_times = [None] * loops
theirs_times = [None] * loops
binbounds = [[0, 1, 2, 3, 4] for x in range(ndim)]
# weights = numpy.random.rand(npts)
weights = numpy.ones((npts,), dtype=numpy.float64)
print('binbounds: {}'.format(binbounds))
print('weights')
print(weights)
for n in range(loops):
testdat = numpy.require(numpy.random.randint(0, 4, size=(npts, ndim)), numpy.uint16)
print('test data')
print(testdat)
mstart = time()
mine = histnd(testdat, binbounds, weights=weights)
mstop = time()
tstart = time()
theirs = numpy.histogramdd(testdat, binbounds, weights=weights)[0]
tstop = time()
mine_times[n] = mstop - mstart
theirs_times[n] = tstop - tstart
print(mine)
print(theirs)
errsum = numpy.abs(mine - theirs).sum()
errsum_per_item = errsum / npts
rel_err = errsum / numpy.abs(weights).sum()
print('sum of the absolute errors: {} ({} relative, {} per entry)'.format(errsum, rel_err, errsum_per_item))
print('mine, best of {}: {}'.format(loops, min(mine_times)))
print('theirs, best of {}: {}'.format(loops, min(theirs_times)))
def _test_double(npts=1024 * 1024, loops=3):
from time import time
mine_times = [None] * loops
theirs_times = [None] * loops
# though 1.0 should be sufficient, weirdness in the boundary conditions
# for numpy.digitize appears to introduce a discrepancy, so throw something
# greater than 1.0 in for good measure
binbounds = [[0, 0.5, 1, 1.1] for x in range(3)]
weights = numpy.random.rand(npts)
# weights = numpy.ones((npts,), dtype=numpy.float64)
for n in range(loops):
testdat = numpy.random.rand(npts, 3)
mstart = time()
mine = histnd(testdat, binbounds, weights=weights)
mstop = time()
tstart = time()
theirs = numpy.histogramdd(testdat, binbounds, weights=weights)[0]
tstop = time()
mine_times[n] = mstop - mstart
theirs_times[n] = tstop - tstart
print(mine)
print(theirs)
errsum = numpy.abs(mine - theirs).sum()
errsum_per_item = errsum / npts
rel_err = errsum / numpy.abs(weights).sum()
print('sum of the absolute errors: {} ({} relative, {} per entry)'.format(errsum, rel_err, errsum_per_item))
print('mine, best of {}: {}'.format(loops, min(mine_times)))
print('theirs, best of {}: {}'.format(loops, min(theirs_times)))
def _remote_bin_iter(iiter, n_iter, dsspec, wt_dsspec, initpoint, binbounds, ignore_out_of_range):
iter_hist_shape = tuple(len(bounds)-1 for bounds in binbounds)
iter_hist = numpy.zeros(iter_hist_shape, dtype=numpy.float64)
dset = dsspec.get_iter_data(n_iter)
npts = dset.shape[1]
weights = wt_dsspec.get_iter_data(n_iter)
dset = dset[:,initpoint:,:]
for ipt in xrange(npts-initpoint):
histnd(dset[:,ipt,:], binbounds, weights, out=iter_hist, binbound_check = False, ignore_out_of_range=ignore_out_of_range)
del weights, dset
# normalize histogram
normhistnd(iter_hist,binbounds)
return iiter, n_iter, iter_hist
n_segs = phis.shape[0]
# stack the phi, psi, and Ntwid (or N) vals for each seg into single array
vals = np.array([phis, psis, nregs]).T
assert vals.shape == (n_segs, 3)
denom = np.dot(phi_vals[:, np.newaxis], ntwids[np.newaxis, :])
assert denom.shape == (n_windows, n_segs)
denom -= free_energies[:, np.newaxis]
denom = np.exp(-denom)
denom *= n_tots[:, np.newaxis]
denom = np.sum(denom, axis=0)
histnd(vals, binbounds, n_segs * weights / denom, out=hist, binbound_check=False)
for dm in data_managers:
dm.close_backing()
# normalize the f****r
normhistnd(hist, binbounds)
phi_binctrs = phi_binbounds[:-1] + np.diff(phi_binbounds) / 2.0
psi_binctrs = psi_binbounds[:-1] + np.diff(psi_binbounds) / 2.0
vmin, vmax = 0, 10
norm = matplotlib.colors.Normalize(vmin=vmin, vmax=vmax)
extent = (phi_binbounds[0], phi_binbounds[-1], psi_binbounds[0], psi_binbounds[-1])
min_energy = 20.0