RHOBINS = np.arange(RHOMIN, RHOMAX+0.1, 0.1) # left edges + rightmost edge
# build corresponding lists of recs and stranges, even entries are desynched, odd are synched:
recs, stranges = [], [] # recs has repetitions, not unique
for rec in urecs: # iterate over sorted unique recs
tranges = REC2STATETRANGES[rec.absname]
for trange in tranges:
recs.append(rec)
stranges.append(trange)
nrecsec = len(recs)
assert nrecsec == len(stranges)
assert nrecsec % 2 == 0 # should always be an even number of recording sections
states = ['desynch', 'synch'] * (nrecsec // 2) # alternating states
# get active or all neuron ids for each section of each recording:
recsecnids = core.get_ssnids(recs, stranges, kind=NIDSKIND)[1]
# calculate responsive PSTHs:
rts = {} # index into by [rec.absname][statei]
rpsths = {} # index into by [rec.absname][statei][nid]
for rec, nids, strange, state in zip(recs, recsecnids, stranges, states):
print(rec.absname, state)
statei = {'desynch': 0, 'synch': 1}[state]
if rec.absname not in rts:
rts[rec.absname] = [None, None] # init, index into using statei
rpsths[rec.absname] = [{}, {}] # init, index into using statei
t, psths, spikets = rec.psth(nids=nids, natexps=False, blank=BLANK, strange=strange,
plot=False, binw=BINW, tres=TRES, gauss=GAUSS, norm='ntrials')
rts[rec.absname][statei] = t
for nid, psth, ts in zip(nids, psths, spikets):
# run PSTH peak detection:
ptc22tr1r08s = [ptc22.tr1.r08, ptc22.tr1.r08]
strangesr08s = [(0, 1500e6), # r08 desynched, us
(1550e6, np.inf)] # r08 synched, us, end is ~ 2300s
ptc22tr1r10s = [ptc22.tr1.r10, ptc22.tr1.r10]
strangesr10s = [(0, 1400e6), # r10 synched, us
(1480e6, np.inf)] # r10 desynched, us, end is ~ 2300s
ptc22tr1recs = [ptc22.tr1.r05, ptc22.tr1.r08, ptc22.tr1.r10, ptc22.tr1.r19]
ptc22tr2recs = [ptc22.tr2.r28, ptc22.tr2.r33]
# ptc15.tr7c:
SEPMAX = 1675
# get superset of active nids for all natexps of both recs in ptc15tr7crecs:
stranges = etrangesr74 + etrangesr95b # 8 stranges in total
recs = [ptc15.tr7c.r74]*4 + [ptc15.tr7c.r95b]*4 # 8 recs corresponding to 8 stranges
ssnids, recsecnids = get_ssnids(recs, stranges)
ssseps = get_seps(ssnids, ptc15.tr7c.alln)
# get separate supersets of active nids for all 4 natexps in each recording:
ptc15tr7crecsecnids = [np.unique(np.hstack(recsecnids[:4])),
np.unique(np.hstack(recsecnids[4:]))]
# do psthcorr plots and collect ssrho matrices:
ssrhos = []
for rec, nids in zip(ptc15tr7crecs, ptc15tr7crecsecnids):
ssrho = psthcorr(rec, nids=nids, ssnids=ssnids, ssseps=ssseps, natexps=True) # in sec
ssrhos.append(ssrho)
# plot differences in superset rho matrices for the two recordings:
psthcorrdiff(ssrhos, ssseps, 'r74-r95b')
## rho for ns1 figure:
#In [124]: np.where(ssnids == 328)
#Out[124]: (array([31]),)
def psthcorrtype(trackrecs, pool=False, alpha=ALPHA1, vmin=VMIN, vmax=VMAX,
separatetypeplots=True):
"""Plot mean PSTH correlation (rho) 2D histograms, indexed by spike and RF type. Plot one
for each set of recordings in trackrecs (ostensibly, one per track). If pool, plot only
one rho celltype histogram pooled across all trackrecs. If pool, return rhotype matrix
filled only with those distributions whose mean is significantly different from 0."""
ntracks = len(trackrecs)
tracknames = [ trackrec[0].tr.absname for trackrec in trackrecs ]
rhotype = listarr(np.empty((8, 8))) # init rho cell type 2D array of lists
npairs = 0 # init npairs
for tracki, recs in enumerate(trackrecs):
track = recs[0].tr
natexps = False
trackname = recs[0].tr.absname
if trackname == 'ptc15.tr7c':
natexps = True
ssnids, recsecnids = get_ssnids(recs)
ssrhos = []
for rec in recs:
ssrho = psthcorr(rec, nids=None, ssnids=ssnids, natexps=natexps, plot=False)
ssrhos.append(ssrho)
ssrhos = np.asarray(ssrhos) # convert to 3D array
if pool == False:
listarr(rhotype) # reset between tracks
npairs = 0 # reset between tracks
nn = len(ssnids)
nanis = np.isnan(ssrhos) # indices of nan values
ssrhos[nanis] = 0 # replace nans with 0s
maxabsssrhos = core.maxabs(ssrhos, axis=0) # keep only the max rho of each cell pair
alln = track.alln
for i in range(nn):
ni = alln[ssnids[i]] # neuron i
si = celltype2int[ni.spiketype]
ri = celltype2int[ni.rftype]
for j in range(i+1, nn): # use only upper triangle, don't double count cell pairs
nj = alln[ssnids[j]] # neuron j
sj = celltype2int[nj.spiketype]
rj = celltype2int[nj.rftype]
rho = maxabsssrhos[i, j]
if rho == 0:
# ignore this cell pair's rho (they were never simultaneously active) so it
# doesn't mess up the celltype stats
continue
# fill in the upper triangle of rhotype matrix:
rhotype[si, sj].append(rho)
rhotype[ri, rj].append(rho)
# these cross terms should best be left disabled, because they conflate the
# correlations between spiketype and rftype:
#rhotype[ri, sj].append(rho)
#rhotype[si, rj].append(rho)
npairs += 1
rhotypemeans = np.zeros(rhotype.shape); rhotypemeans.fill(nan)
rhotypestds = np.zeros(rhotype.shape); rhotypestds.fill(nan)
rhotypeps = np.zeros(rhotype.shape); rhotypeps.fill(np.inf)
rhotypesigmeans = np.zeros(rhotype.shape); rhotypesigmeans.fill(nan)
# calculate rho stats for each combination of cell type:
for i in range(8):
for j in range(i, 8): # use only upper triangle, don't double count celltype stats
if len(rhotype[i, j]) > 1:
rhotypemeans[i, j] = np.mean(rhotype[i, j])
rhotypestds[i, j] = np.std(rhotype[i, j])
# 2-sided sample mean ttest relative to 0:
t, p = ttest_1samp(rhotype[i, j], 0)
rhotypeps[i, j] = p
sigis = rhotypeps < alpha # indices of significant deviations of mean from 0
rhotypesigmeans[sigis] = rhotypemeans[sigis]
#arrs = [rhotypemeans, rhotypestds, rhotypeps, rhotypesigmeans]
#plottypes = ['mean', 'stdev', 'pval', 'sigmean']
arrs = [rhotypesigmeans]
plottypes = ['sigmean']
if pool:
if tracki < ntracks-1:
continue # only plot once all tracks have been iterated over
trackname = ', '.join(tracknames)
for arr, plottype in zip(arrs, plottypes):
# get symmetric arr by copying upper triangle, transposing to get lower triangle,
# and adding to arr:
symarr = nansum([arr, np.triu(arr, k=1).T], axis=0)
thisvmin, thisvmax = nanmin(symarr), nanmax(symarr)
vmin = min(vmin, thisvmin) # set to manual vmin at most
vmax = max(vmax, thisvmax) # set to manual vmax at least
if separatetypeplots:
figure(figsize=(8, 3))
# plot spiketypes:
subplot(121)
imshow(symarr[:4, :4], vmin=vmin, vmax=vmax, origin='upper', cmap='jet')
xticks(np.arange(4), spiketypelabels, rotation=90)
yticks(np.arange(4), spiketypelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
# plot rftypes:
subplot(122)
imshow(symarr[4:, 4:], vmin=vmin, vmax=vmax, origin='upper', cmap='jet')
xticks(np.arange(4), rftypelabels, rotation=90)
yticks(np.arange(4), rftypelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
plottype += ' separate'
else: # plot spike and rf types in the same matrix
figure(figsize=(4, 4))
imshow(symarr, vmin=vmin, vmax=vmax, origin='upper', cmap='jet')
xticks(np.arange(8), typelabels, rotation=90)
yticks(np.arange(8), typelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
plottype += ' combined'
titlestr = (trackname + ' psthcorrtype ' + plottype +
' alpha=%.4f, npairs=%d' % (alpha, npairs))
gcfm().window.setWindowTitle(titlestr)
tight_layout(pad=0.4)
if pool: # only return rhotype if pooling across all tracks
insigis = np.logical_not(sigis)
rhotype[insigis] = listarr(rhotype[insigis]) # set insig entries to empty lists
return rhotype # only significant entires aren't empty
# build corresponding lists of recs and stranges, even entries are desynched, odd are synched:
recs, stranges = [], [] # recs has repetitions, not unique
for rec in urecs: # iterate over sorted unique recs
tranges = REC2STATETRANGES[rec.absname]
for trange in tranges: # desynched, then synched
recs.append(rec)
stranges.append(trange)
nrecsec = len(recs)
assert nrecsec == len(stranges)
assert nrecsec % 2 == 0 # should always be an even number of recording sections
stateis = [0, 1] * (nrecsec // 2) # allternating desynched and synched state indices
fmts = ['b-', 'r-'] * (nrecsec // 2) # alternating plotting formats for desynched and synched
# get active or all neuron ids for each section of each recording:
recsecnids = get_ssnids(recs, stranges, kind='all')[1]
## TODO: this loop is a mess. It should have 3 nested loops (recording, state, neuron)
## instead of the current 2 (recording*state, neuron). That's why it has the messy
## "% 2 == 0" stuff. There's no need to use the get_ssnids() function if kind='all',
## which it always is.
# calculate PSTHs for all sections of all recordings, collect data from each PSTH with at
# least 1 detected peak:
psthss, spiketss = [], []
rpsths = [[], []] # responsive PSTHs, indexed by state
psthparamsrecsec = [] # params returned for each PSTH, for each recording section
widthsrecsec = [] # peak widths, for each recording section
tsrecsec = [] # peak times, for each recording section
heightsrecsec = [] # peak heights, for each recording section
peaknspikesrecsec = [] # spike counts of each peak, normalized by ntrials, for each rec section
RHOBINS = np.arange(RHOMIN, RHOMAX + 0.1, 0.1) # left edges + rightmost edge
# build corresponding lists of recs and stranges, even entries are desynched, odd are synched:
recs, stranges = [], [] # recs has repetitions, not unique
for rec in urecs: # iterate over sorted unique recs
tranges = REC2STATETRANGES[rec.absname]
for trange in tranges:
recs.append(rec)
stranges.append(trange)
nrecsec = len(recs)
assert nrecsec == len(stranges)
assert nrecsec % 2 == 0 # should always be an even number of recording sections
states = ['desynch', 'synch'] * (nrecsec // 2) # alternating states
# get active or all neuron ids for each section of each recording:
recsecnids = core.get_ssnids(recs, stranges, kind=NIDSKIND)[1]
# calculate responsive PSTHs:
rts = {} # index into by [rec.absname][statei]
rpsths = {} # index into by [rec.absname][statei][nid]
for rec, nids, strange, state in zip(recs, recsecnids, stranges, states):
print(rec.absname, state)
statei = {'desynch': 0, 'synch': 1}[state]
if rec.absname not in rts:
rts[rec.absname] = [None, None] # init, index into using statei
rpsths[rec.absname] = [{}, {}] # init, index into using statei
t, psths, spikets = rec.psth(nids=nids,
natexps=False,
blank=BLANK,
strange=strange,
plot=False,
recs, stranges = [], [] # recs has repetitions, not unique
for rec in urecs: # iterate over sorted unique recs
tranges = REC2STATETRANGES[rec.absname]
for trange in tranges: # desynched, then synched
recs.append(rec)
stranges.append(trange)
nrecsec = len(recs)
assert nrecsec == len(stranges)
assert nrecsec % 2 == 0 # should always be an even number of recording sections
stateis = [0, 1] * (nrecsec // 2
) # allternating desynched and synched state indices
fmts = ['b-', 'r-'] * (
nrecsec // 2) # alternating plotting formats for desynched and synched
# get active or all neuron ids for each section of each recording:
recsecnids = get_ssnids(recs, stranges, kind='all')[1]
## TODO: this loop is a mess. It should have 3 nested loops (recording, state, neuron)
## instead of the current 2 (recording*state, neuron). That's why it has the messy
## "% 2 == 0" stuff. There's no need to use the get_ssnids() function if kind='all',
## which it always is.
# calculate PSTHs for all sections of all recordings, collect data from each PSTH with at
# least 1 detected peak:
psthss, spiketss = [], []
rpsths = [[], []] # responsive PSTHs, indexed by state
psthparamsrecsec = [
] # params returned for each PSTH, for each recording section
widthsrecsec = [] # peak widths, for each recording section
tsrecsec = [] # peak times, for each recording section
heightsrecsec = [] # peak heights, for each recording section
def psthcorrtype(trackrecs,
pool=False,
alpha=ALPHA1,
vmin=VMIN,
vmax=VMAX,
separatetypeplots=True):
"""Plot mean PSTH correlation (rho) 2D histograms, indexed by spike and RF type. Plot one
for each set of recordings in trackrecs (ostensibly, one per track). If pool, plot only
one rho celltype histogram pooled across all trackrecs. If pool, return rhotype matrix
filled only with those distributions whose mean is significantly different from 0."""
ntracks = len(trackrecs)
tracknames = [trackrec[0].tr.absname for trackrec in trackrecs]
rhotype = listarr(np.empty((8, 8))) # init rho cell type 2D array of lists
npairs = 0 # init npairs
for tracki, recs in enumerate(trackrecs):
track = recs[0].tr
natexps = False
trackname = recs[0].tr.absname
if trackname == 'ptc15.tr7c':
natexps = True
ssnids, recsecnids = get_ssnids(recs)
ssrhos = []
for rec in recs:
ssrho = psthcorr(rec,
nids=None,
ssnids=ssnids,
natexps=natexps,
plot=False)
ssrhos.append(ssrho)
ssrhos = np.asarray(ssrhos) # convert to 3D array
if pool == False:
listarr(rhotype) # reset between tracks
npairs = 0 # reset between tracks
nn = len(ssnids)
nanis = np.isnan(ssrhos) # indices of nan values
ssrhos[nanis] = 0 # replace nans with 0s
maxabsssrhos = core.maxabs(
ssrhos, axis=0) # keep only the max rho of each cell pair
alln = track.alln
for i in range(nn):
ni = alln[ssnids[i]] # neuron i
si = celltype2int[ni.spiketype]
ri = celltype2int[ni.rftype]
for j in range(
i + 1, nn
): # use only upper triangle, don't double count cell pairs
nj = alln[ssnids[j]] # neuron j
sj = celltype2int[nj.spiketype]
rj = celltype2int[nj.rftype]
rho = maxabsssrhos[i, j]
if rho == 0:
# ignore this cell pair's rho (they were never simultaneously active) so it
# doesn't mess up the celltype stats
continue
# fill in the upper triangle of rhotype matrix:
rhotype[si, sj].append(rho)
rhotype[ri, rj].append(rho)
# these cross terms should best be left disabled, because they conflate the
# correlations between spiketype and rftype:
#rhotype[ri, sj].append(rho)
#rhotype[si, rj].append(rho)
npairs += 1
rhotypemeans = np.zeros(rhotype.shape)
rhotypemeans.fill(nan)
rhotypestds = np.zeros(rhotype.shape)
rhotypestds.fill(nan)
rhotypeps = np.zeros(rhotype.shape)
rhotypeps.fill(np.inf)
rhotypesigmeans = np.zeros(rhotype.shape)
rhotypesigmeans.fill(nan)
# calculate rho stats for each combination of cell type:
for i in range(8):
for j in range(
i, 8
): # use only upper triangle, don't double count celltype stats
if len(rhotype[i, j]) > 1:
rhotypemeans[i, j] = np.mean(rhotype[i, j])
rhotypestds[i, j] = np.std(rhotype[i, j])
# 2-sided sample mean ttest relative to 0:
t, p = ttest_1samp(rhotype[i, j], 0)
rhotypeps[i, j] = p
sigis = rhotypeps < alpha # indices of significant deviations of mean from 0
rhotypesigmeans[sigis] = rhotypemeans[sigis]
#arrs = [rhotypemeans, rhotypestds, rhotypeps, rhotypesigmeans]
#plottypes = ['mean', 'stdev', 'pval', 'sigmean']
arrs = [rhotypesigmeans]
plottypes = ['sigmean']
if pool:
if tracki < ntracks - 1:
continue # only plot once all tracks have been iterated over
trackname = ', '.join(tracknames)
for arr, plottype in zip(arrs, plottypes):
# get symmetric arr by copying upper triangle, transposing to get lower triangle,
# and adding to arr:
symarr = nansum([arr, np.triu(arr, k=1).T], axis=0)
thisvmin, thisvmax = nanmin(symarr), nanmax(symarr)
vmin = min(vmin, thisvmin) # set to manual vmin at most
vmax = max(vmax, thisvmax) # set to manual vmax at least
if separatetypeplots:
figure(figsize=(8, 3))
# plot spiketypes:
subplot(121)
imshow(symarr[:4, :4],
vmin=vmin,
vmax=vmax,
origin='upper',
cmap='jet')
xticks(np.arange(4), spiketypelabels, rotation=90)
yticks(np.arange(4), spiketypelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
# plot rftypes:
subplot(122)
imshow(symarr[4:, 4:],
vmin=vmin,
vmax=vmax,
origin='upper',
cmap='jet')
xticks(np.arange(4), rftypelabels, rotation=90)
yticks(np.arange(4), rftypelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
plottype += ' separate'
else: # plot spike and rf types in the same matrix
figure(figsize=(4, 4))
imshow(symarr,
vmin=vmin,
vmax=vmax,
origin='upper',
cmap='jet')
xticks(np.arange(8), typelabels, rotation=90)
yticks(np.arange(8), typelabels)
colorbar(ticks=(vmin, vmax), format='%.2f')
plottype += ' combined'
titlestr = (trackname + ' psthcorrtype ' + plottype +
' alpha=%.4f, npairs=%d' % (alpha, npairs))
gcfm().window.setWindowTitle(titlestr)
tight_layout(pad=0.4)
if pool: # only return rhotype if pooling across all tracks
insigis = np.logical_not(sigis)
rhotype[insigis] = listarr(
rhotype[insigis]) # set insig entries to empty lists
return rhotype # only significant entires aren't empty
ptc22tr1r10s = [ptc22.tr1.r10, ptc22.tr1.r10]
strangesr10s = [
(0, 1400e6), # r10 synched, us
(1480e6, np.inf)
] # r10 desynched, us, end is ~ 2300s
ptc22tr1recs = [ptc22.tr1.r05, ptc22.tr1.r08, ptc22.tr1.r10, ptc22.tr1.r19]
ptc22tr2recs = [ptc22.tr2.r28, ptc22.tr2.r33]
# ptc15.tr7c:
SEPMAX = 1675
# get superset of active nids for all natexps of both recs in ptc15tr7crecs:
stranges = etrangesr74 + etrangesr95b # 8 stranges in total
recs = [ptc15.tr7c.r74] * 4 + [ptc15.tr7c.r95b
] * 4 # 8 recs corresponding to 8 stranges
ssnids, recsecnids = get_ssnids(recs, stranges)
ssseps = get_seps(ssnids, ptc15.tr7c.alln)
# get separate supersets of active nids for all 4 natexps in each recording:
ptc15tr7crecsecnids = [
np.unique(np.hstack(recsecnids[:4])),
np.unique(np.hstack(recsecnids[4:]))
]
# do psthcorr plots and collect ssrho matrices:
ssrhos = []
for rec, nids in zip(ptc15tr7crecs, ptc15tr7crecsecnids):
ssrho = psthcorr(rec,
nids=nids,
ssnids=ssnids,
ssseps=ssseps,
natexps=True) # in sec
ssrhos.append(ssrho)
