1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
|
import os
from os import path as op
import numpy as np
from numpy.polynomial import legendre
from ..parallel import parallel_func
from ..utils import logger, _get_extra_data_path
##############################################################################
# FAST LEGENDRE (DERIVATIVE) POLYNOMIALS USING LOOKUP TABLE
def _next_legen_der(n, x, p0, p01, p0d, p0dd):
"""Compute the next Legendre polynomial and its derivatives"""
# only good for n > 1 !
help_ = p0
helpd = p0d
p0 = ((2 * n - 1) * x * help_ - (n - 1) * p01) / n
p0d = n * help_ + x * helpd
p0dd = (n + 1) * helpd + x * p0dd
p01 = help_
return p0, p0d, p0dd
def _get_legen(x, n_coeff=100):
"""Get Legendre polynomials expanded about x"""
return legendre.legvander(x, n_coeff - 1)
def _get_legen_der(xx, n_coeff=100):
"""Get Legendre polynomial derivatives expanded about x"""
coeffs = np.empty((len(xx), n_coeff, 3))
for c, x in zip(coeffs, xx):
p0s, p0ds, p0dds = c[:, 0], c[:, 1], c[:, 2]
p0s[:2] = [1.0, x]
p0ds[:2] = [0.0, 1.0]
p0dds[:2] = [0.0, 0.0]
for n in range(2, n_coeff):
p0s[n], p0ds[n], p0dds[n] = _next_legen_der(n, x, p0s[n - 1],
p0s[n - 2], p0ds[n - 1],
p0dds[n - 1])
return coeffs
def _get_legen_table(ch_type, volume_integral=False, n_coeff=100,
n_interp=20000, force_calc=False):
"""Return a (generated) LUT of Legendre (derivative) polynomial coeffs"""
if n_interp % 2 != 0:
raise RuntimeError('n_interp must be even')
fname = op.join(_get_extra_data_path(), 'tables')
if not op.isdir(fname):
# Updated due to API chang (GH 1167)
os.makedirs(fname)
if ch_type == 'meg':
fname = op.join(fname, 'legder_%s_%s.bin' % (n_coeff, n_interp))
leg_fun = _get_legen_der
extra_str = ' derivative'
lut_shape = (n_interp + 1, n_coeff, 3)
else: # 'eeg'
fname = op.join(fname, 'legval_%s_%s.bin' % (n_coeff, n_interp))
leg_fun = _get_legen
extra_str = ''
lut_shape = (n_interp + 1, n_coeff)
if not op.isfile(fname) or force_calc:
n_out = (n_interp // 2)
logger.info('Generating Legendre%s table...' % extra_str)
x_interp = np.arange(-n_out, n_out + 1, dtype=np.float64) / n_out
lut = leg_fun(x_interp, n_coeff).astype(np.float32)
if not force_calc:
with open(fname, 'wb') as fid:
fid.write(lut.tostring())
else:
logger.info('Reading Legendre%s table...' % extra_str)
with open(fname, 'rb', buffering=0) as fid:
lut = np.fromfile(fid, np.float32)
lut.shape = lut_shape
# we need this for the integration step
n_fact = np.arange(1, n_coeff, dtype=float)
if ch_type == 'meg':
n_facts = list() # multn, then mult, then multn * (n + 1)
if volume_integral:
n_facts.append(n_fact / ((2.0 * n_fact + 1.0)
* (2.0 * n_fact + 3.0)))
else:
n_facts.append(n_fact / (2.0 * n_fact + 1.0))
n_facts.append(n_facts[0] / (n_fact + 1.0))
n_facts.append(n_facts[0] * (n_fact + 1.0))
# skip the first set of coefficients because they are not used
lut = lut[:, 1:, [0, 1, 1, 2]] # for multiplicative convenience later
# reshape this for convenience, too
n_facts = np.array(n_facts)[[2, 0, 1, 1], :].T
n_facts = np.ascontiguousarray(n_facts)
n_fact = n_facts
else: # 'eeg'
n_fact = (2.0 * n_fact + 1.0) * (2.0 * n_fact + 1.0) / n_fact
# skip the first set of coefficients because they are not used
lut = lut[:, 1:].copy()
return lut, n_fact
def _get_legen_lut_fast(x, lut):
"""Return Legendre coefficients for given x values in -1<=x<=1"""
# map into table vals (works for both vals and deriv tables)
n_interp = (lut.shape[0] - 1.0)
# equiv to "(x + 1.0) / 2.0) * n_interp" but faster
mm = x * (n_interp / 2.0) + 0.5 * n_interp
# nearest-neighbor version (could be decent enough...)
idx = np.round(mm).astype(int)
vals = lut[idx]
return vals
def _get_legen_lut_accurate(x, lut):
"""Return Legendre coefficients for given x values in -1<=x<=1"""
# map into table vals (works for both vals and deriv tables)
n_interp = (lut.shape[0] - 1.0)
# equiv to "(x + 1.0) / 2.0) * n_interp" but faster
mm = x * (n_interp / 2.0) + 0.5 * n_interp
# slower, more accurate interpolation version
mm = np.minimum(mm, n_interp - 0.0000000001)
idx = np.floor(mm).astype(int)
w2 = mm - idx
w2.shape += tuple([1] * (lut.ndim - w2.ndim)) # expand to correct size
vals = (1 - w2) * lut[idx] + w2 * lut[idx + 1]
return vals
def _comp_sum_eeg(beta, ctheta, lut_fun, n_fact):
"""Lead field dot products using Legendre polynomial (P_n) series"""
# Compute the sum occurring in the evaluation.
# The result is
# sums[:] (2n+1)^2/n beta^n P_n
coeffs = lut_fun(ctheta)
betans = np.cumprod(np.tile(beta[:, np.newaxis], (1, n_fact.shape[0])),
axis=1)
s0 = np.dot(coeffs * betans, n_fact) # == weighted sum across cols
return s0
def _comp_sums_meg(beta, ctheta, lut_fun, n_fact, volume_integral):
"""Lead field dot products using Legendre polynomial (P_n) series"""
# Compute the sums occurring in the evaluation.
# Two point magnetometers on the xz plane are assumed.
# The four sums are:
# * sums[:, 0] n(n+1)/(2n+1) beta^(n+1) P_n
# * sums[:, 1] n/(2n+1) beta^(n+1) P_n'
# * sums[:, 2] n/((2n+1)(n+1)) beta^(n+1) P_n'
# * sums[:, 3] n/((2n+1)(n+1)) beta^(n+1) P_n''
coeffs = lut_fun(ctheta)
beta = (np.cumprod(np.tile(beta[:, np.newaxis], (1, n_fact.shape[0])),
axis=1) * beta[:, np.newaxis])
# This is equivalent, but slower:
# sums = np.sum(beta[:, :, np.newaxis] * n_fact * coeffs, axis=1)
# sums = np.rollaxis(sums, 2)
sums = np.einsum('ij,jk,ijk->ki', beta, n_fact, coeffs)
return sums
###############################################################################
# SPHERE DOTS
def _fast_sphere_dot_r0(r, rr1, rr2, lr1, lr2, cosmags1, cosmags2,
w1, w2, volume_integral, lut, n_fact, ch_type):
"""Lead field dot product computation for M/EEG in the sphere model"""
ct = np.einsum('ik,jk->ij', rr1, rr2) # outer product, sum over coords
# expand axes
rr1 = rr1[:, np.newaxis, :] # (n_rr1, n_rr2, n_coord) e.g. 4x4x3
rr2 = rr2[np.newaxis, :, :]
lr1lr2 = lr1[:, np.newaxis] * lr2[np.newaxis, :]
beta = (r * r) / lr1lr2
if ch_type == 'meg':
sums = _comp_sums_meg(beta.flatten(), ct.flatten(), lut, n_fact,
volume_integral)
sums.shape = (4,) + beta.shape
# Accumulate the result, a little bit streamlined version
#cosmags1 = cosmags1[:, np.newaxis, :]
#cosmags2 = cosmags2[np.newaxis, :, :]
#n1c1 = np.sum(cosmags1 * rr1, axis=2)
#n1c2 = np.sum(cosmags1 * rr2, axis=2)
#n2c1 = np.sum(cosmags2 * rr1, axis=2)
#n2c2 = np.sum(cosmags2 * rr2, axis=2)
#n1n2 = np.sum(cosmags1 * cosmags2, axis=2)
n1c1 = np.einsum('ik,ijk->ij', cosmags1, rr1)
n1c2 = np.einsum('ik,ijk->ij', cosmags1, rr2)
n2c1 = np.einsum('jk,ijk->ij', cosmags2, rr1)
n2c2 = np.einsum('jk,ijk->ij', cosmags2, rr2)
n1n2 = np.einsum('ik,jk->ij', cosmags1, cosmags2)
part1 = ct * n1c1 * n2c2
part2 = n1c1 * n2c1 + n1c2 * n2c2
result = (n1c1 * n2c2 * sums[0] +
(2.0 * part1 - part2) * sums[1] +
(n1n2 + part1 - part2) * sums[2] +
(n1c2 - ct * n1c1) * (n2c1 - ct * n2c2) * sums[3])
# Give it a finishing touch!
const = 4e-14 * np.pi # This is \mu_0^2/4\pi
result *= (const / lr1lr2)
if volume_integral:
result *= r
else: # 'eeg'
sums = _comp_sum_eeg(beta.flatten(), ct.flatten(), lut, n_fact)
sums.shape = beta.shape
# Give it a finishing touch!
eeg_const = 1.0 / (4.0 * np.pi)
result = eeg_const * sums / lr1lr2
# new we add them all up with weights
if w1 is None: # operating on surface, treat independently
#result = np.sum(w2[np.newaxis, :] * result, axis=1)
result = np.dot(result, w2)
else:
#result = np.sum((w1[:, np.newaxis] * w2[np.newaxis, :]) * result)
result = np.einsum('i,j,ij', w1, w2, result)
return result
def _do_self_dots(intrad, volume, coils, r0, ch_type, lut, n_fact, n_jobs):
"""Perform the lead field dot product integrations"""
if ch_type == 'eeg':
intrad *= 0.7
# convert to normalized distances from expansion center
rmags = [coil['rmag'] - r0[np.newaxis, :] for coil in coils]
rlens = [np.sqrt(np.sum(r * r, axis=1)) for r in rmags]
rmags = [r / rl[:, np.newaxis] for r, rl in zip(rmags, rlens)]
cosmags = [coil['cosmag'] for coil in coils]
ws = [coil['w'] for coil in coils]
parallel, p_fun, _ = parallel_func(_do_self_dots_subset, n_jobs)
prods = parallel(p_fun(intrad, rmags, rlens, cosmags,
ws, volume, lut, n_fact, ch_type, idx)
for idx in np.array_split(np.arange(len(rmags)), n_jobs))
products = np.sum(prods, axis=0)
return products
def _do_self_dots_subset(intrad, rmags, rlens, cosmags, ws, volume, lut,
n_fact, ch_type, idx):
"""Helper for parallelization"""
products = np.zeros((len(rmags), len(rmags)))
for ci1 in idx:
for ci2 in range(0, ci1 + 1):
res = _fast_sphere_dot_r0(intrad, rmags[ci1], rmags[ci2],
rlens[ci1], rlens[ci2],
cosmags[ci1], cosmags[ci2],
ws[ci1], ws[ci2], volume, lut,
n_fact, ch_type)
products[ci1, ci2] = res
products[ci2, ci1] = res
return products
def _do_surface_dots(intrad, volume, coils, surf, sel, r0, ch_type,
lut, n_fact, n_jobs):
"""Compute the map construction products"""
virt_ref = False
# convert to normalized distances from expansion center
rmags = [coil['rmag'] - r0[np.newaxis, :] for coil in coils]
rlens = [np.sqrt(np.sum(r * r, axis=1)) for r in rmags]
rmags = [r / rl[:, np.newaxis] for r, rl in zip(rmags, rlens)]
cosmags = [coil['cosmag'] for coil in coils]
ws = [coil['w'] for coil in coils]
rref = None
refl = None
if ch_type == 'eeg':
intrad *= 0.7
if virt_ref:
rref = virt_ref[np.newaxis, :] - r0[np.newaxis, :]
refl = np.sqrt(np.sum(rref * rref, axis=1))
rref /= refl[:, np.newaxis]
rsurf = surf['rr'][sel] - r0[np.newaxis, :]
lsurf = np.sqrt(np.sum(rsurf * rsurf, axis=1))
rsurf /= lsurf[:, np.newaxis]
this_nn = surf['nn'][sel]
parallel, p_fun, _ = parallel_func(_do_surface_dots_subset, n_jobs)
prods = parallel(p_fun(intrad, rsurf, rmags, rref, refl, lsurf, rlens,
this_nn, cosmags, ws, volume, lut, n_fact, ch_type,
idx)
for idx in np.array_split(np.arange(len(rmags)), n_jobs))
products = np.sum(prods, axis=0)
return products
def _do_surface_dots_subset(intrad, rsurf, rmags, rref, refl, lsurf, rlens,
this_nn, cosmags, ws, volume, lut, n_fact, ch_type,
idx):
"""Helper for parallelization"""
products = np.zeros((len(rsurf), len(rmags)))
for ci in idx:
res = _fast_sphere_dot_r0(intrad, rsurf, rmags[ci],
lsurf, rlens[ci],
this_nn, cosmags[ci],
None, ws[ci], volume, lut,
n_fact, ch_type)
if rref is not None:
vres = _fast_sphere_dot_r0(intrad, rref, rmags[ci],
refl, rlens[ci],
None, ws[ci], volume,
lut, n_fact, ch_type)
products[:, ci] = res - vres
else:
products[:, ci] = res
return products
|