def rrps(cells):
"""
cells: CellDoc -> (cns:N-[ of s, sns:N-[ of s,
hists:{cond1 cond2:N,N-# of x})
cns is a list of cell names, sns is a list of stimulus names, hists is
a dictionary of arrays such that hists[c][i,j] is the fraction of total
responses of cell cns[i] in condition c which occured in respons to stimulus
sns[j].
"""
cns = cells.keys(0, 'cell', sort=True)
stims = stimnames(cells[cns[0]])
hists = {'cond1': np.zeros((len(cns), len(stims))),
'cond2': np.zeros((len(cns), len(stims)))}
for i, cn in enumerate(cns):
for cond in ['cond1', 'cond2']:
nts = len(flat(cells[cn][cond]['evts']))
if not nts:
continue
for sno, sname in enumerate(stims):
evts = rfromstim(cells[cn], cond, sno)
if evts:
rr = float(len(flat(evts))) / float(nts)
else:
rr = 0
hists[cond][i, sno] = rr
#hists[cond][i,:] = np.argsort(hists[cond][i,:])
return (cns, stims, hists)
def showav(d, fig=1, yps=50, btop=100000, bw=2000):
bins = np.arange(0, btop, bw)
f = plt.figure(fig)
plt.clf()
cnds = dconds(d)
n = len(cnds)
yl = 0
for i, c in enumerate(cnds):
sp = plt.subplot(1, n, i + 1)
sp.xaxis.set_visible(False)
if i > 0:
sp.yaxis.set_visible(False)
sp.xaxis.set_visible(False)
plt.title(c)
l = len(d[c]['evts'])
for j in range(l):
evts = d[c]['evts'][j]
if evts:
x = np.array(evts)
y = np.zeros_like(x) + j
plt.plot(x, y, marker='.', color='r', linestyle='None')
z = nhplt(flat(d[c]['evts']), bins, color='r', bottom=l + .2 * yps)
yl = max(yl, l + 1.3 * yps)
if d[c]['avgs']:
z = nhplt(flat(d[c]['avgs']), bins, color='b', bottom=l + .1 * yps)
if d[c]['dj']:
for j in range(l):
evts = d[c]['dj'][j]
if evts:
x = np.array(evts)
y = np.zeros_like(x) + j
plt.plot(x, y, marker='.', color='g', linestyle='None')
z = nhplt(flat(d[c]['dj']), bins, color='g', bottom=l + .1 * yps)
if d[c]['avg'] != None:
avl = int(l / 2.0)
if len(d[c]['avg']) == 0:
plt.axhspan(avl, avl + 1, color='b')
else:
x = np.array(d[c]['avg'])
y = np.zeros_like(x) + avl
plt.plot(x, y, marker='o', color='b', markersize=10.0, linestyle='None')
if d[c]['avgvar'] != None:
av = d[c]['avgvar']
a = [0] + list(x) + [max(flat(d[c]['evts']))]
nx = [(a[i] + a[i - 1]) / 2.0 for i in range(1, len(a))]
ny = np.zeros_like(nx) + avl
ye = ([0] * len(av), [v[4] * yps for v in av])
plt.errorbar(nx, ny, yerr=ye, color='k', marker='.',
markersize=4.0, linestyle='None', elinewidth=3)
av = av[:-1]
xmc = np.array([v[1] for v in av])
xe = [v[2] for v in av]
ymc = np.array([v[3] * yps for v in av])
plt.errorbar(x + xmc, y + ymc, xerr=xe, marker='s', color='b',
markersize=6.0, linestyle='None', elinewidth=3)
for i in range(len(cnds)):
sp = plt.subplot(1, n, i + 1)
plt.ylim([0, yl])
f.canvas.draw()
def sfate(evts, q, tst=None, track=None, near=False):
ef = flat(evts)
tfs = []
if tst:
for e in evts:
tfs.extend(vpd_trans(e, tst, q))
ntrans = len(evts)
holes = [0] * (len(tst) + 1)
else:
n = len(evts)
ntrans = 0
for i in range(n - 1):
for j in range(i, n):
ntrans += 1
tfs.extend(vpd_trans(evts[i], evts[j], q))
holes = False
if track == None:
if tst:
track = tst
else:
track = sorted(set(ef))
elif type(track) == int:
ef = flat(evts)
track = np.linspace(min(ef), max(ef), track)
track = list(track)
fates = [[] for _ in range(len(track))]
for tf in tfs:
if near:
tf = list(tf)
for i in range(len(tf)):
if tf[i] != -1:
tf[i] = track[nearest(track, [tf[i]])[0]]
if tf[0] == -1:
try:
id = track.index(tf[1])
fates[id].append(np.inf)
except ValueError:
pass
elif tf[1] == -1:
try:
id = track.index(tf[0])
fates[id].append(-np.inf)
except ValueError:
if holes:
id = np.searchsorted(track, tf[0])
holes[id] += 1
else:
try:
id = track.index(tf[0])
fates[id].append(tf[1] - tf[0])
except ValueError:
pass
try:
id = track.index(tf[1])
fates[id].append(tf[0] - tf[1])
except ValueError:
pass
return (track, fates, ntrans, holes)
def srate(d):
rate = float(len(flat(d['evts']))) / len(d['evts'])
mrate = rate
st = np.array(d['stims'])
for s in np.unique(st):
evts = [d['evts'][i] for i in np.nonzero(st == s)[0]]
r = float(len(flat(evts))) / len(evts)
mrate = max(r, mrate)
return {'max_rate': mrate, 'mean_rate': rate}
def addpois(d, conds, l=200):
td = gd.Doc()
for k in conds:
td[k] = d[k]
arate = float(len(flat(d[k]['evts']))) / len(d[k]['evts'])
arate = arate / (l / 1e6)
#report (arate)
stims = d[k]['stims']
mi = AC(l, arate)
td[k + '_hp'] = {'stims': stims, 'evts': [mi(st) for st in stims]}
td[k + '_hp.rate'] = arate
td['docsource'] = d['docsource']
return td
#
def histpl(d, k, bw=4.0, rng=(0, 200.0)):
evts = d[k + '.evts']
evts = np.array(flat(evts))
evts = evts / 1000.0
nb = (rng[1] - rng[0]) / bw
plt.hist(evts, nb, rng)
def _rpp(d):
npres = 0
nspks = 0
for k in d:
if d[k]['evts']:
npres += len(d[k]['evts'])
nspks += len(flat(d[k]['evts']))
if npres == 0:
return -1
return float(nspks) / float(npres)
def optq(evts, mode='med-vdps', reps=1, minq=1000, qstep=1000):
q = minq - qstep
expt = round(float(len(flat(evts))) / len(evts))
nm = 0
while nm < expt:
print(nm, expt, q)
q = q + qstep
m = calcav(evts, mode, q, reps, True)
nm = len(m)
return q
def rburst(n=200, prate=10):
dur = 100000
d = gd.Doc()
mp = T.AC(dur, prate, 1, 3)
d['condpois'] = T.draw_single(mp, 0, n)
m = T.BURST(dur, 1, 15000, 1, sd=1000, latsd=2000, mfst=10000, sprob=.8, slp=2, mpt=3)
b = T.draw_single(m, 0, n)
rt = 1e6 * len(flat(b['evts'])) / float(len(b['evts'])) / dur
print('burst1 rate', rt)
p = T.draw_single(mp, 0, n)
rt = 1e6 * len(flat(p['evts'])) / float(len(p['evts'])) / dur
print('pois rate', rt)
d['condb1'] = combinedraws(b, p)
m = T.BURST(dur, 1, 10000, 1, sd=2000, latsd=2000, mfst=15000, sprob=.9, slp=3, mpt=2)
b = T.draw_single(m, 0, n)
rt = 1e6 * len(flat(b['evts'])) / float(len(b['evts'])) / dur
print('burst2 rate', rt)
p = T.draw_single(mp, 0, n)
d['condb2'] = combinedraws(b, p)
return d
def spikes_from(cond, stim, flatten=True):
"""
return a flat list of every spike in cond['evts'] resulting from stmulus
stim (int).
"""
resps = [cond['evts'][i] for i in range(len(cond['evts']))
if cond['stims'][i] == stim]
if flatten:
return flat(resps)
else:
return resps
def ll_et(evts, bw=2000):
nt = len(evts)
evts = flat(evts)
spp = float(len(evts)) / nt
bins = cover(evts, bw)
h = np.histogram(evts, bins=bins, normed=True)
h0 = h[0] * spp * bw
zi = np.searchsorted(h[1], evts, 'right') - 1
# for i in range(len(h0)):
# print(h[0][i], (zi==i).sum())
ll2 = np.log(h0[zi]).sum()
return (ll2 / nt, (h0, h[1]))
def nburst(n=100, rt=15.0, dur=100, prec=15, sd=2000, latsd=5000, slp=2, mpt=3, sprob=.9):
dur = int(dur * 1000)
prec = int(prec * 1000)
m = T.BURST(dur, rt, prec, 1, sd=sd, latsd=latsd, sprob=sprob, slp=slp, mpt=mpt)
d = gd.Doc()
d['condburst'] = T.draw_single(m, 0, n)
rt = d['condburst.evts']
rt = len(flat(rt)) / float(len(rt))
rt = rt * 1e6 / dur
print(rt)
m = T.AC(dur, rt, prec, 3)
d['condhpois'] = T.draw_single(m, 1, n)
return d
def dj_hist(evts, q, bw=2000, nsteps=4, insp=0):
ll, sph = ll_et(evts, bw)
ll = ll / float(len(flat(evts)))
while True:
start = sph[1][0]
end = sph[1][-1]
tp = transprob_hist(evts, q, bw)
nevts = []
for e in evts:
ne = []
for st in e:
bid = nearest(tp[:, 0], [st])[0]
sm, ssd, dp = tp[bid, 1:-1]
#be careful, if the implementation of "cover" used by transprob_hist and
#ll_et isn't exactly the same, this doesn't work
sp = sph[0][bid]
#check for deletion
if (1 - sp) * dp * insp > sp:
continue
shifts = np.arange(-nsteps, nsteps + 1) * bw
shifts = np.array([s for s in shifts if s + st > start and s + st < end])
lc = [hdraw(sph, st + s) * gausact(sm, ssd, s, bw) for s in shifts]
opts = shifts[np.argmax(lc)]
ne.append(st + opts)
nevts.append(sorted(ne))
nnspk = len(flat(nevts))
if not nnspk:
break
nll, sph = ll_et(nevts, bw)
nll = nll / nnspk
if nll <= ll:
break
evts = nevts
print(ll, '->', nll)
ll = nll
return evts
def superpose(io, n, israte=False, ct=.3):
if israte:
rt = float(len(flat(io['evts']))) / len(io['evts'])
n, rem = divmod(n / rt, 1)
if rem > ct:
n += 1
if n < 2:
return io
ss = stimsets(io)
stims = []
evts = []
for s in ss:
for g in groupevts(ss[s], n):
stims.append(s)
evts.append(g)
return {'stims': stims, 'evts': evts, 'ncomb': n}
def run(self, pars, out, messages):
d = pars['doc']
xvar = pars['xvar']
r = pars['r']
out.patch(gd.Doc({'xvar': xvar, 'docsource': d['docsource']}))
for cond in d.keys(0, 'cond'):
out[cond] = {'nspikes': len(flat(d[cond]['evts'])),
'npres': len(d[cond]['evts'])}
sc = r.findall({'_params.io': '=->%s' % cond})
if sc:
out[cond]['params'] = r[sc[0]]['_params']
m = {}
for k in sc:
x = r[k]['_params'][xvar]
m[x] = r[k]['mi'] + [r[k]['stiment']]
out[cond]['x'] = np.array(sorted(m))
out[cond]['y'] = np.array([m[i] for i in out[cond]['x']])
def response_densities(cond, mcent=8, nrep=3, stims=None,
sppt=163000, dsd=10000):
"""
return a list l such that l[i] = is a gmm of the responses to stimulus i.
Since these are 1D models, the underlying mixmod call always uses PkLkCk,
but mcent specifies the max number of centers to try, nrep the number of
repeats to use.
stims may be a list, which restricts the set of stimuli that are modeled
(by default, it is all presented stimuli).
If the number of spikes evoked by a stimulus is very small, mixmod errors
will result. This function will not try to calculate a model based on fewer
than 2 spikes per center, reguardless of the value of mcent. For response
groups with at least two spikes, mcent may be reduced. Responses with no
spikes are modeled with a 0-component model (the dictionary
{'components':(), 'support':sppt}. mixmod.evaluate on such a model will
return a uniform probability if 1/sppt, so this behavior is equivalent to a
uniform prior over a region of size sppt. Responses with exactly 1 spike are
modeled as a single center, with mean at the time of that spike, and
standard deviation specified by the free parameter dsd.
This function adds the key "responses" to the resulting model dictionaries,
containing the source spike trains
"""
if stims is None:
stims = set(cond['stims'])
l = [None for _ in range(max(stims) + 1)]
for s in stims:
sts = spikes_from(cond, s, False)
sf = flat(sts)
if len(sf) > 1:
mc = int(min(mcent, np.ceil(len(sf) / 2.0)))
l[s] = mmcall(np.array(sf)[:, np.newaxis], range(1, mc + 1),
reps=nrep)
elif sf:
l[s] = {'proportions': (1.0,), 'c0': {'cov': ((dsd ** 2,),),
'mean': (sf[0],)},
'bic': 0, 'components': ('c0',), 'partition': (0,)}
else:
l[s] = {'components': (), 'support': sppt}
l[s]['responses'] = sts
return l
def showpartitions(parts, evts, av="med-vdps", aq=15000, hists=True):
'''
Plot one or more partitions of a set of events. These are plotted as rasters,
with events in each group of the partition shown together in a given color.
Different partitions are shown in different subplots. "parts" should be a
list of arrays, each specifing a partition. "evts" is a list of event
sequences, av may be a False value, or a method string accepted by calcav. If
it is specified, each group in the partition is also represented by an
average calculated using this method, and "aq" as the precision parameter. If
"hists" is true, the groups are represented with historgrams as well as
rasters
'''
f = plt.figure(1)
plt.clf()
for pi, part in enumerate(parts):
plt.subplot(1, len(parts), pi + 1)
pids = np.unique(part)
y = 0
pcols = {}
for i, pid in enumerate(pids):
pcols[pid] = (COLORS[i], y)
pevts = [evts[i] for i in range(len(evts)) if part[i] == pid]
for j, e in enumerate(pevts):
x = np.array(e)
yc = np.zeros_like(x) + y
plt.plot(x, yc, marker='.', color=pcols[pid][0], linestyle='None')
y += 1
if av:
yav = float(pcols[pid][1] + y) / 2
x = calcav(pevts, av, aq, 1, True)
if len(x) == 0:
plt.axhspan(yav - .2, yav + .2, color=pcols[pid][0])
else:
x = np.array(x)
yc = np.zeros_like(x) + yav
plt.plot(x, yc, marker='o', color=pcols[pid][0], markersize=10.0, linestyle='None')
yz = 0
if hists:
for pid in pids:
pevts = [evts[i] for i in range(len(evts)) if part[i] == pid]
z = nhplt(flat(pevts), 50, color=pcols[pid][0], bottom=y + 1, scale=y)
yz = max(z, yz)
plt.ylim([-1, y + 2 + yz])
f.canvas.draw()
def addtests(d, l=200, match='cond1', q=10000, mods=(AC, TGC), sp={}):
td = gd.Doc()
iod = d[match]
arate = float(len(flat(iod['evts']))) / len(iod['evts'])
arate = arate / (l / 1e6)
report(arate)
stims = iod['stims']
nstims = np.unique(stims).shape[0]
l = int(l)
td[match] = iod
for k in d:
if k.startswith('cond') and k != match:
td[k] = d[k]
for M in mods:
kw = sp.get(M.name, {})
mi = M(l, arate, q, nstims, **kw)
td['cond' + M.name] = {'stims': stims, 'evts': [mi(st) for st in stims]}
td['docsource'] = {'cell': d['docsource.cell'] + '+tests',
'pars': d['docsource.pars'] + (l, match, q)}
return td
def transprob_hist(evts, q, bw=2000):
bins = cover(flat(evts), bw)
n = len(evts)
nt = (n ** 2 - n) / 2
tfs = []
for i in range(n - 1):
for j in range(i, n):
tfs.append(vpd_trans(evts[i], evts[j], q))
fates = []
for _ in range(len(bins)):
fates.append(np.zeros((len(tfs), 2)))
for i, trans in enumerate(tfs):
for tf in trans:
if tf[0] == -1:
id = nearest(bins, [tf[1]])[0]
fates[id][i, 1] = 1
elif tf[1] == -1:
id = nearest(bins, [tf[0]])[0]
fates[id][i, 1] = 1
else:
id1 = nearest(bins, [tf[0]])[0]
id2 = nearest(bins, [tf[1]])[0]
dist = tf[1] - tf[0]
fates[id1][i, 0] = dist
fates[id2][i, 0] = -dist
res = np.zeros((len(bins), 5))
for i in range(len(bins)):
res[i, 0] = bins[i]
ns = float((fates[i][:, 0] != 0).sum() + (fates[i][:, 1] != 0).sum())
res[i, 4] = ns / nt
v = [x for x in fates[i][:, 0] if x != 0]
if len(v) < ns:
v = v + [0] * int(ns - len(v))
v = np.array(v)
res[i, 1] = v.mean()
res[i, 2] = v.std()
res[i, 3] = fates[i][:, 1].sum() / ns
return res
def tr(l=100000, r=1.0):
a = AC(l, r)
evts = [a(0) for i in range(1000)]
report(len(flat(evts)) / 1000.0)
