ts[solver][i] = Timer(
lambda: foopsi(y, g=[g], lam=2.4, solver=solver)).timeit(
number=runs) / runs
except SolverError:
print("The solver " + solver +
" is actually not installed, hence skipping it.")
break
constrained_ts = {}
for solver in solvers[:-1]: # GUROBI failed
constrained_ts[solver] = np.nan * np.zeros(N)
print(
'running %7s with p=1 and optimizing lambda such that noise constraint is tight'
% solver)
for i, y in enumerate(Y):
if solver == 'OASIS':
constrained_ts[solver][i] = Timer(lambda: constrained_oasisAR1(
y, g=g, sn=sn)).timeit(number=runs) / runs
else:
try:
constrained_ts[solver][i] = Timer(lambda: constrained_foopsi(
y, g=[g], sn=sn, solver=solver)).timeit(number=runs) / runs
except SolverError:
print("The solver " + solver +
" is actually not installed, hence skipping it.")
break
constrained_ts['GUROBI'] = np.zeros(N) * np.nan # GUROBI failed
# plot
fig = plt.figure(figsize=(7, 5))
fig.add_axes([.14, .17, .79, .82])
plt.errorbar(range(len(solvers)), [np.mean(ts[s]) for s in solvers],
[np.std(ts[s]) / np.sqrt(N) for s in solvers],
def deconvolve(y,
g=(None, ),
sn=None,
b=None,
optimize_g=0,
penalty=0,
**kwargs):
"""Infer the most likely discretized spike train underlying an fluorescence trace
Solves the noise constrained sparse non-negative deconvolution problem
min |s|_q subject to |c-y|^2 = sn^2 T and s = Gc >= 0
where q is either 1 or 0, rendering the problem convex or non-convex.
Parameters:
-----------
y : array, shape (T,)
Fluorescence trace.
g : tuple of float, optional, default (None,)
Parameters of the autoregressive model, cardinality equivalent to p.
Estimated from the autocovariance of the data if no value is given.
sn : float, optional, default None
Standard deviation of the noise distribution. If no value is given,
then sn is estimated from the data based on power spectral density if not provided.
b : float, optional, default None
Fluorescence baseline value. If no value is given, then b is optimized.
optimize_g : int, optional, default 0
Number of large, isolated events to consider for optimizing g.
If optimize_g=0 the provided or estimated g is not further optimized.
penalty : int, optional, default 1
Sparsity penalty. 1: min |s|_1 0: min |s|_0
kwargs : dict
Further keywords passed on to constrained_oasisAR1 or constrained_onnlsAR2.
Returns:
--------
c : array, shape (T,)
The inferred denoised fluorescence signal at each time-bin.
s : array, shape (T,)
Discretized deconvolved neural activity (spikes).
b : float
Fluorescence baseline value.
g : tuple of float
Parameters of the AR(2) process that models the fluorescence impulse response.
lam: float
Optimal Lagrange multiplier for noise constraint under L1 penalty
"""
if g[0] is None or sn is None:
est = estimate_parameters(y, p=len(g), fudge_factor=.98)
if g[0] is None:
g = est[0]
if sn is None:
sn = est[1]
if len(g) == 1:
return constrained_oasisAR1(y,
g[0],
sn,
optimize_b=True if b is None else False,
optimize_g=optimize_g,
penalty=penalty,
**kwargs)
elif len(g) == 2:
if optimize_g > 0:
raise NotImplementedError(
'Optimization of AR parameters currenty only supported for AR(1)'
)
return constrained_onnlsAR2(y,
g,
sn,
optimize_b=True if b is None else False,
optimize_g=optimize_g,
penalty=penalty,
**kwargs)
else:
print 'g must have length 1 or 2, cause only AR(1) and AR(2) are currently implemented'
def constrained_onnlsAR2(y,
g,
sn,
optimize_b=True,
optimize_g=0,
decimate=5,
shift=100,
window=200,
tol=1e-9,
max_iter=1,
penalty=1):
""" Infer the most likely discretized spike train underlying an AR(2) fluorescence trace
Solves the noise constrained sparse non-negative deconvolution problem
min |s|_1 subject to |c-y|^2 = sn^2 T and s_t = c_t-g1 c_{t-1}-g2 c_{t-2} >= 0
Parameters
----------
y : array of float
One dimensional array containing the fluorescence intensities (with baseline
already subtracted) with one entry per time-bin.
g : (float, float)
Parameters of the AR(2) process that models the fluorescence impulse response.
sn : float
Standard deviation of the noise distribution.
optimize_b : bool, optional, default True
Optimize baseline if True else it is set to 0, see y.
optimize_g : int, optional, default 0
Number of large, isolated events to consider for optimizing g.
No optimization if optimize_g=0.
decimate : int, optional, default 5
Decimation factor for estimating hyper-parameters faster on decimated data.
max_iter : int, optional, default 1
Maximal number of iterations.
penalty : int, optional, default 1
Sparsity penalty. 1: min |s|_1 0: min |s|_0
Returns
-------
c : array of float
The inferred denoised fluorescence signal at each time-bin.
s : array of float
Discretized deconvolved neural activity (spikes).
b : float
Fluorescence baseline value.
(g1, g2) : tuple of float
Parameters of the AR(2) process that models the fluorescence impulse response.
lam : float
Sparsity penalty parameter lambda of dual problem.
References
----------
* Friedrich J and Paninski L, NIPS 2016
* Friedrich J, Zhou P, and Paninski L, arXiv 2016
"""
T = len(y)
d = (g[0] + sqrt(g[0] * g[0] + 4 * g[1])) / 2
r = (g[0] - sqrt(g[0] * g[0] + 4 * g[1])) / 2
if not optimize_g:
g11 = (np.exp(log(d) * np.arange(1, T + 1)) -
np.exp(log(r) * np.arange(1, T + 1))) / (d - r)
g12 = np.append(0, g[1] * g11[:-1])
g11g11 = np.cumsum(g11 * g11)
g11g12 = np.cumsum(g11 * g12)
Sg11 = np.cumsum(g11)
f_lam = 1 - g[0] - g[1]
thresh = sn * sn * T
# get initial estimate of b and lam on downsampled data using AR1 model
if decimate > 0:
_, s, b, aa, lam = constrained_oasisAR1(y.reshape(-1,
decimate).mean(1),
d**decimate,
sn / sqrt(decimate),
optimize_b=optimize_b,
optimize_g=optimize_g)
if optimize_g > 0:
d = aa**(1. / decimate)
g[0] = d + r
g[1] = -d * r
g11 = (np.exp(log(d) * np.arange(1, T + 1)) -
np.exp(log(r) * np.arange(1, T + 1))) / (d - r)
g12 = np.append(0, g[1] * g11[:-1])
g11g11 = np.cumsum(g11 * g11)
g11g12 = np.cumsum(g11 * g12)
Sg11 = np.cumsum(g11)
f_lam = 1 - g[0] - g[1]
lam *= (1 - d**decimate) / f_lam
ff = np.hstack([
a * decimate + np.arange(-decimate, decimate)
for a in np.where(s > 1e-6)[0]
]) # this window size seems necessary and sufficient
ff = np.unique(ff[(ff >= 0) * (ff < T)])
mask = np.zeros(T, dtype=bool)
mask[ff] = True
else:
b = np.percentile(y, 15) if optimize_b else 0
lam = 2 * sn * np.linalg.norm(g11)
mask = None
# run ONNLS
c, s = onnls(y - b, g, lam=lam, mask=mask)
if not optimize_b: # don't optimize b, just the dual variable lambda
for i in range(max_iter - 1):
res = y - c
RSS = res.dot(res)
if np.abs(RSS - thresh) < 1e-4:
break
# calc shift dlam, here attributed to sparsity penalty
tmp = np.empty(T)
ls = np.append(np.where(s > 1e-6)[0], T)
l = ls[0]
tmp[:l] = (1 + d) / (1 + d**l) * np.exp(
log(d) * np.arange(l)) # first pool
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f - 1
# if and elif correct last 2 time points for |s|_1 instead |c|_1
if i == len(ls) - 2: # last pool
tmp[f] = (1. / f_lam if l == 0 else
(Sg11[l] + g[1] / f_lam * g11[l - 1] +
(g[0] + g[1]) / f_lam * g11[l] -
g11g12[l] * tmp[f - 1]) / g11g11[l])
# secondlast pool if last one has length 1
elif i == len(ls) - 3 and ls[-2] == T - 1:
tmp[f] = (Sg11[l] + g[1] / f_lam * g11[l] -
g11g12[l] * tmp[f - 1]) / g11g11[l]
else: # all other pools
tmp[f] = (Sg11[l] - g11g12[l] * tmp[f - 1]) / g11g11[l]
l += 1
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
aa = tmp.dot(tmp)
bb = res.dot(tmp)
cc = RSS - thresh
try:
db = (-bb + sqrt(bb * bb - aa * cc)) / aa
except:
os.write(1, 'shit happens\n')
db = -bb / aa
# perform shift
b += db
c, s = onnls(y - b, g, lam=lam, mask=mask)
db = np.mean(y - c) - b
b += db
lam -= db / f_lam
else: # optimize b
db = np.mean(y - c) - b
b += db
lam -= db / (1 - g[0] - g[1])
for i in range(max_iter - 1):
res = y - c - b
RSS = res.dot(res)
if np.abs(RSS - thresh) < 1e-4:
break
# calc shift db, here attributed to baseline
tmp = np.empty(T)
ls = np.append(np.where(s > 1e-6)[0], T)
l = ls[0]
tmp[:l] = (1 + d) / (1 + d**l) * np.exp(
log(d) * np.arange(l)) # first pool
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f
tmp[f] = (Sg11[l - 1] -
g11g12[l - 1] * tmp[f - 1]) / g11g11[l - 1]
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
tmp -= tmp.mean()
aa = tmp.dot(tmp)
bb = res.dot(tmp)
cc = RSS - thresh
try:
db = (-bb + sqrt(bb * bb - aa * cc)) / aa
except:
os.write(1, 'shit happens\n')
db = -bb / aa
# perform shift
b += db
c, s = onnls(y - b, g, lam=lam, mask=mask)
db = np.mean(y - c) - b
b += db
lam -= db / f_lam
if penalty == 0: # get (locally optimal) L0 solution
def c4smin(y, s, s_min):
ls = np.append(np.where(s > s_min)[0], T)
tmp = np.zeros_like(s)
l = ls[0] # first pool
tmp[:l] = max(
0,
np.exp(log(d) * np.arange(l)).dot(y[:l]) * (1 - d * d) /
(1 - d**(2 * l))) * np.exp(log(d) * np.arange(l))
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f
tmp[f] = (g11[:l].dot(y[f:f + l]) -
g11g12[l - 1] * tmp[f - 1]) / g11g11[l - 1]
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
return tmp
spikesizes = np.sort(s[s > 1e-6])
i = len(spikesizes) / 2
l = 0
u = len(spikesizes) - 1
while u - l > 1:
s_min = spikesizes[i]
tmp = c4smin(y - b, s, s_min)
res = y - b - tmp
RSS = res.dot(res)
if RSS < thresh or i == 0:
l = i
i = (l + u) / 2
res0 = tmp
else:
u = i
i = (l + u) / 2
if i > 0:
c = res0
s = np.append([0, 0], c[2:] - g[0] * c[1:-1] - g[1] * c[:-2])
return c, s, b, g, lam
def constrained_onnlsAR2(y,
g,
sn,
optimize_b=True,
b_nonneg=True,
optimize_g=0,
decimate=5,
shift=100,
window=None,
tol=1e-9,
max_iter=1,
penalty=1):
""" Infer the most likely discretized spike train underlying an AR(2) fluorescence trace
Solves the noise constrained sparse non-negative deconvolution problem
min |s|_1 subject to |c-y|^2 = sn^2 T and s_t = c_t-g1 c_{t-1}-g2 c_{t-2} >= 0
Parameters
----------
y : array of float
One dimensional array containing the fluorescence intensities (with baseline
already subtracted) with one entry per time-bin.
g : (float, float)
Parameters of the AR(2) process that models the fluorescence impulse response.
sn : float
Standard deviation of the noise distribution.
optimize_b : bool, optional, default True
Optimize baseline if True else it is set to 0, see y.
b_nonneg: bool, optional, default True
Enforce strictly non-negative baseline if True.
optimize_g : int, optional, default 0
Number of large, isolated events to consider for optimizing g.
No optimization if optimize_g=0.
decimate : int, optional, default 5
Decimation factor for estimating hyper-parameters faster on decimated data.
shift : int, optional, default 100
Number of frames by which to shift window from on run of NNLS to the next.
window : int, optional, default None (200 or larger dependend on g)
Window size.
tol : float, optional, default 1e-9
Tolerance parameter.
max_iter : int, optional, default 1
Maximal number of iterations.
penalty : int, optional, default 1
Sparsity penalty. 1: min |s|_1 0: min |s|_0
Returns
-------
c : array of float
The inferred denoised fluorescence signal at each time-bin.
s : array of float
Discretized deconvolved neural activity (spikes).
b : float
Fluorescence baseline value.
(g1, g2) : tuple of float
Parameters of the AR(2) process that models the fluorescence impulse response.
lam : float
Sparsity penalty parameter lambda of dual problem.
References
----------
* Friedrich J and Paninski L, NIPS 2016
* Friedrich J, Zhou P, and Paninski L, PLOS Computational Biology 2017
"""
T = len(y)
d = (g[0] + sqrt(g[0] * g[0] + 4 * g[1])) / 2
r = (g[0] - sqrt(g[0] * g[0] + 4 * g[1])) / 2
if window is None:
window = int(min(T, max(200, -5 / log(d))))
if not optimize_g:
g11 = (np.exp(log(d) * np.arange(1, T + 1)) * np.arange(1, T + 1)) \
if d == r else \
(np.exp(log(d) * np.arange(1, T + 1)) -
np.exp(log(r) * np.arange(1, T + 1))) / (d - r)
g12 = np.append(0, g[1] * g11[:-1])
g11g11 = np.cumsum(g11 * g11)
g11g12 = np.cumsum(g11 * g12)
Sg11 = np.cumsum(g11)
f_lam = 1 - g[0] - g[1]
elif decimate == 0: # need to run AR1 anyways for estimating AR coeffs
decimate = 1
thresh = sn * sn * T
# get initial estimate of b and lam on downsampled data using AR1 model
if decimate > 0:
_, s, b, aa, lam = constrained_oasisAR1(
y[:len(y) // decimate * decimate].reshape(-1, decimate).mean(1),
d**decimate,
sn / sqrt(decimate),
optimize_b=optimize_b,
b_nonneg=b_nonneg,
optimize_g=optimize_g)
if optimize_g:
d = aa**(1. / decimate)
if decimate > 1:
s = oasisAR1(y - b, d, lam=lam * (1 - aa) / (1 - d))[1]
r = estimate_time_constant(s, 1, fudge_factor=.98)[0]
g[0] = d + r
g[1] = -d * r
g11 = (np.exp(log(d) * np.arange(1, T + 1)) -
np.exp(log(r) * np.arange(1, T + 1))) / (d - r)
g12 = np.append(0, g[1] * g11[:-1])
g11g11 = np.cumsum(g11 * g11)
g11g12 = np.cumsum(g11 * g12)
Sg11 = np.cumsum(g11)
f_lam = 1 - g[0] - g[1]
elif decimate > 1:
s = oasisAR1(y - b, d, lam=lam * (1 - aa) / (1 - d))[1]
lam *= (1 - d**decimate) / f_lam
# s = oasisAR1(s, r)[1]
# this window size seems necessary and sufficient
ff = np.ravel(
[a + np.arange(-2, 2) for a in np.where(s > s.max() / 10.)[0]])
ff = np.unique(ff[(ff >= 0) * (ff < T)]).astype(int)
mask = np.zeros(T, dtype=bool)
mask[ff] = True
else:
b = np.percentile(y, 15) if optimize_b else 0
lam = 2 * sn * np.linalg.norm(g11)
mask = None
if b_nonneg:
b = max(b, 0)
# run ONNLS
c, s = onnls(y - b,
g,
lam=lam,
mask=mask,
shift=shift,
window=window,
tol=tol)
g_converged = False
if not optimize_b: # don't optimize b, just the dual variable lambda and g if optimize_g
for i in range(max_iter - 1):
res = y - c
RSS = res.dot(res)
if np.abs(RSS - thresh) < 1e-4:
break
# calc shift dlam, here attributed to sparsity penalty
tmp = np.empty(T)
ls = np.append(np.where(s > 1e-6)[0], T)
l = ls[0]
tmp[:l] = (1 + d) / (1 + d**l) * np.exp(
log(d) * np.arange(l)) # first pool
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f - 1
# if and elif correct last 2 time points for |s|_1 instead |c|_1
if i == len(ls) - 2: # last pool
tmp[f] = (1. / f_lam if l == 0 else
(Sg11[l] + g[1] / f_lam * g11[l - 1] +
(g[0] + g[1]) / f_lam * g11[l] -
g11g12[l] * tmp[f - 1]) / g11g11[l])
# secondlast pool if last one has length 1
elif i == len(ls) - 3 and ls[-2] == T - 1:
tmp[f] = (Sg11[l] + g[1] / f_lam * g11[l] -
g11g12[l] * tmp[f - 1]) / g11g11[l]
else: # all other pools
tmp[f] = (Sg11[l] - g11g12[l] * tmp[f - 1]) / g11g11[l]
l += 1
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
aa = tmp.dot(tmp)
bb = res.dot(tmp)
cc = RSS - thresh
try:
dlam = (-bb + sqrt(bb * bb - aa * cc)) / aa
except:
dlam = -bb / aa
# perform shift
lam += dlam / f_lam
c, s = onnls(y,
g,
lam=lam,
mask=mask,
shift=shift,
window=window,
tol=tol)
# update g
if optimize_g and (not g_converged):
lengths = np.where(s)[0][1:] - np.where(s)[0][:-1]
def getRSS(y, opt):
ld, lr = opt
if ld < lr:
return 1e3 * thresh
d, r = exp(ld), exp(lr)
g1, g2 = d + r, -d * r
tmp = onnls(y, [g1, g2], lam,
mask=(s > 1e-2 * s.max()))[0] - y
return tmp.dot(tmp)
result = minimize(lambda x: getRSS(y, x), (log(d), log(r)),
bounds=((None, -1e-4), (None, -1e-3)),
method='L-BFGS-B',
options={
'gtol': 1e-04,
'maxiter': 10,
'ftol': 1e-05
})
if abs(result['x'][1] - log(d)) < 1e-4:
g_converged = True
ld, lr = result['x']
d, r = exp(ld), exp(lr)
g = (d + r, -d * r)
c, s = onnls(y,
g,
lam=lam,
mask=mask,
shift=shift,
window=window,
tol=tol)
else: # optimize b
db = max(np.mean(y - c), 0 if b_nonneg else -np.inf) - b
b += db
lam -= db / (1 - g[0] - g[1])
for i in range(max_iter - 1):
res = y - c - b
RSS = res.dot(res)
if np.abs(RSS - thresh) < 1e-4:
break
# calc shift db, here attributed to baseline
tmp = np.empty(T)
ls = np.append(np.where(s > 1e-6)[0], T)
l = ls[0]
tmp[:l] = (1 + d) / (1 + d**l) * np.exp(
log(d) * np.arange(l)) # first pool
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f
tmp[f] = (Sg11[l - 1] -
g11g12[l - 1] * tmp[f - 1]) / g11g11[l - 1]
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
tmp -= tmp.mean()
aa = tmp.dot(tmp)
bb = res.dot(tmp)
cc = RSS - thresh
try:
db = (-bb + sqrt(bb * bb - aa * cc)) / aa
except:
db = -bb / aa
# perform shift
if b_nonneg:
db = max(db, -b)
b += db
c, s = onnls(y - b,
g,
lam=lam,
mask=mask,
shift=shift,
window=window,
tol=tol)
# update b and lam
db = max(np.mean(y - c), 0 if b_nonneg else -np.inf) - b
b += db
lam -= db / f_lam
# update g and b
if optimize_g and (not g_converged):
lengths = np.where(s)[0][1:] - np.where(s)[0][:-1]
def getRSS(y, opt):
b, ld, lr = opt
if ld < lr:
return 1e3 * thresh
d, r = exp(ld), exp(lr)
g1, g2 = d + r, -d * r
tmp = b + onnls(
y - b, [g1, g2], lam, mask=(s > 1e-2 * s.max()))[0] - y
return tmp.dot(tmp)
result = minimize(lambda x: getRSS(y, x), (b, log(d), log(r)),
bounds=((0 if b_nonneg else None, None),
(None, -1e-4), (None, -1e-3)),
method='L-BFGS-B',
options={
'gtol': 1e-04,
'maxiter': 10,
'ftol': 1e-05
})
if abs(result['x'][1] - log(d)) < 1e-3:
g_converged = True
b, ld, lr = result['x']
d, r = exp(ld), exp(lr)
g = (d + r, -d * r)
c, s = onnls(y - b,
g,
lam=lam,
mask=mask,
shift=shift,
window=window,
tol=tol)
# update b and lam
db = max(np.mean(y - c), 0 if b_nonneg else -np.inf) - b
b += db
lam -= db
if penalty == 0: # get (locally optimal) L0 solution
def c4smin(y, s, s_min):
ls = np.append(np.where(s > s_min)[0], T)
tmp = np.zeros_like(s)
l = ls[0] # first pool
tmp[:l] = max(
0,
np.exp(log(d) * np.arange(l)).dot(y[:l]) * (1 - d * d) /
(1 - d**(2 * l))) * np.exp(log(d) * np.arange(l))
for i, f in enumerate(ls[:-1]): # all other pools
l = ls[i + 1] - f
tmp[f] = (g11[:l].dot(y[f:f + l]) -
g11g12[l - 1] * tmp[f - 1]) / g11g11[l - 1]
tmp[f + 1:f + l] = g11[1:l] * tmp[f] + g12[1:l] * tmp[f - 1]
return tmp
spikesizes = np.sort(s[s > 1e-6])
i = len(spikesizes) // 2
l = 0
u = len(spikesizes) - 1
while u - l > 1:
s_min = spikesizes[i]
tmp = c4smin(y - b, s, s_min)
res = y - b - tmp
RSS = res.dot(res)
if RSS < thresh or i == 0:
l = i
i = (l + u) // 2
res0 = tmp
else:
u = i
i = (l + u) // 2
if i > 0:
c = res0
s = np.append([0, 0], c[2:] - g[0] * c[1:-1] - g[1] * c[:-2])
return c, s, b, g, lam
plot_trace(n)
# plot result after rerunning oasis to fix violations
solution, active_set = foo(active_set, g, ll)
ax = fig.add_axes([ax1, .31, 1 - ax1, .12])
plot_trace(n)
# do few more iterations
for _ in range(3):
solution, active_set, lam = update_lam(y, solution, active_set, g, lam,
sn * sn * len(y))
solution, active_set, g = update_g(y, active_set, g, lam)
# plot converged results with comparison traces
ax = fig.add_axes([ax1, .07, 1 - ax1, .12])
sol_given_g = constrained_oasisAR1(y, .95, sn)[0]
estimated_g = estimate_parameters(y, p=1)[0][0]
print('estimated gamma via autocorrelation: ', estimated_g)
print('optimized gamma : ', g)
sol_PSD_g = oasisAR1(y, estimated_g, 0)[0]
# print((sol_PSD_g-y).dot(sol_PSD_g-y), sn*sn*T # renders constraint problem infeasible
plt.plot(sol_given_g, '--', c=col[6], label=r'true $\gamma$', zorder=11)
plt.plot(sol_PSD_g, c=col[5], label=r'$\gamma$ from autocovariance', zorder=10)
plt.legend(frameon=False, loc=(.1, .62), ncol=2)
plot_trace(n)
plt.xticks([300, 600, 900, 1200], [10, 20, 30, 40])
plt.xlabel('Time [s]', labelpad=-10)
plt.show()
print('correlation with ground truth calcium for given gamma ',
np.corrcoef(sol_given_g, trueC[n])[0, 1])
results = {}
for opt in [
'-', 'l', 'lb', 'lbg', 'lbg10', 'lbg5', 'lbg_ds', 'lbg10_ds', 'lbg5_ds'
]:
results[opt] = {}
results[opt]['time'] = []
results[opt]['distance'] = []
results[opt]['correlation'] = []
for i, y in enumerate(Y):
g, sn = estimate_parameters(y, p=1, fudge_factor=.99)
lam = 2 * sn * (1 - g * g)**(-.5)
b = np.percentile(y, 15)
if opt == '-':
foo = lambda y: oasisAR1(y - b, g, lam)
elif opt == 'l':
foo = lambda y: constrained_oasisAR1(y - b, g, sn)
elif opt == 'lb':
foo = lambda y: constrained_oasisAR1(y, g, sn, optimize_b=True)
elif opt == 'lbg':
foo = lambda y: constrained_oasisAR1(
y, g, sn, optimize_b=True, optimize_g=len(y))
elif opt == 'lbg10':
foo = lambda y: constrained_oasisAR1(
y, g, sn, optimize_b=True, optimize_g=10)
elif opt == 'lbg5':
foo = lambda y: constrained_oasisAR1(
y, g, sn, optimize_b=True, optimize_g=5)
elif opt == 'lbg_ds':
foo = lambda y: constrained_oasisAR1(
y, g, sn, optimize_b=True, optimize_g=len(y), decimate=10)
elif opt == 'lbg10_ds':
def denoise(s):
tmp = cse.deconvolution.estimate_parameters(s, 1, fudge_factor=.97)
constrained_oasisAR1(s, tmp[0][0], tmp[1], True)
def constrained_foopsi(fluor, bl=None, c1=None, g=None, sn=None, p=None, method='cvxpy', bas_nonneg=True,
noise_range=[.25, .5], noise_method='logmexp', lags=5, fudge_factor=1.,
verbosity=False, solvers=None, optimize_g=0, penalty=1, **kwargs):
""" Infer the most likely discretized spike train underlying a fluorescence trace
It relies on a noise constrained deconvolution approach
Parameters
----------
fluor: np.ndarray
One dimensional array containing the fluorescence intensities with
one entry per time-bin.
bl: [optional] float
Fluorescence baseline value. If no value is given, then bl is estimated
from the data.
c1: [optional] float
value of calcium at time 0
g: [optional] list,float
Parameters of the AR process that models the fluorescence impulse response.
Estimated from the data if no value is given
sn: float, optional
Standard deviation of the noise distribution. If no value is given,
then sn is estimated from the data.
p: int
order of the autoregression model
method: [optional] string
solution method for basis projection pursuit 'cvx' or 'cvxpy'
bas_nonneg: bool
baseline strictly non-negative
noise_range: list of two elms
frequency range for averaging noise PSD
noise_method: string
method of averaging noise PSD
lags: int
number of lags for estimating time constants
fudge_factor: float
fudge factor for reducing time constant bias
verbosity: bool
display optimization details
solvers: list string
primary and secondary (if problem unfeasible for approx solution) solvers to be used with cvxpy, default is ['ECOS','SCS']
Returns
-------
c: np.ndarray float
The inferred denoised fluorescence signal at each time-bin.
bl, c1, g, sn : As explained above
sp: ndarray of float
Discretized deconvolved neural activity (spikes)
References
----------
* Pnevmatikakis et al. 2016. Neuron, in press, http://dx.doi.org/10.1016/j.neuron.2015.11.037
* Machado et al. 2015. Cell 162(2):338-350
"""
if p is None:
raise Exception("You must specify the value of p")
if g is None or sn is None:
g, sn = estimate_parameters(fluor, p=p, sn=sn, g=g, range_ff=noise_range,
method=noise_method, lags=lags, fudge_factor=fudge_factor)
if p == 0:
c1 = 0
g = np.array(0)
bl = 0
c = np.maximum(fluor, 0)
sp = c.copy()
else:
if method == 'cvx':
c, bl, c1, g, sn, sp = cvxopt_foopsi(
fluor, b=bl, c1=c1, g=g, sn=sn, p=p, bas_nonneg=bas_nonneg, verbosity=verbosity)
elif method == 'cvxpy':
c, bl, c1, g, sn, sp = cvxpy_foopsi(
fluor, g, sn, b=bl, c1=c1, bas_nonneg=bas_nonneg, solvers=solvers)
elif method == 'oasis':
from oasis import constrained_oasisAR1
if p == 1:
if bl is None:
c, sp, bl, g, _ = constrained_oasisAR1(
fluor, g[0], sn, optimize_b=True, b_nonneg=bas_nonneg,
optimize_g=optimize_g, penalty=penalty)
else:
c, sp, _, g, _ = constrained_oasisAR1(
fluor - bl, g[0], sn, optimize_b=False, penalty=1)
c1 = c[0]
# remove intial calcium to align with the other foopsi methods
# it is added back in function constrained_foopsi_parallel of temporal.py
c -= c1 * g**np.arange(len(fluor))
elif p == 2:
if bl is None:
c, sp, bl, g, _ = constrained_oasisAR2(
fluor, g, sn, optimize_b=True, b_nonneg=bas_nonneg,
optimize_g=optimize_g, penalty=penalty)
else:
c, sp, _, g, _ = constrained_oasisAR2(
fluor - bl, g, sn, optimize_b=False, penalty=1)
c1 = c[0]
d = (g[0] + sqrt(g[0] * g[0] + 4 * g[1])) / 2
c -= c1 * d**np.arange(len(fluor))
else:
raise Exception('OASIS is currently only implemented for p=1 and p=2')
g = np.ravel(g)
else:
raise Exception('Undefined Deconvolution Method')
return c, bl, c1, g, sn, sp
