def makeDecision(self, config, save=True):
"""
Function making the decision upon the predicted probas from analyze
method.
Each file is dealt with as a Recording object, created, loaded,
decisions are made, and saved again.
See Recording.makeDecision for more info.
"""
if self._verbatim > 0:
print(
'\n\n *** DATASET ANALYSIS: MAKING DECISION ON PREDICTIONS ***'
)
tStartGlobal = time.time()
for file_path in self.files:
recording = Recording(file_path, config, verbatim=self._verbatim)
recording.load(config)
tStart = time.time()
recording.makeDecision(config)
tEnd = time.time()
if save:
recording.save(config)
if self._verbatim > 1:
print(
'\tRecording has been re-analyzed: decisions on predictions have been made',
file_path, tEnd - tStart)
tEndGlobal = time.time()
if self._verbatim > 0:
print(
'Dataset has been re-analyzed: decisions on predictions have been made',
tEndGlobal - tStartGlobal)
return
def display(self,
config,
onlineDisplay=False,
saveDisplay=True,
forChecking=False,
labelEncoder=None):
"""
Function displaying all files of the dataset.
Each file is dealt with as a Recording object, created and displayed.
See Recording.analyze for more info.
labelEncoder is only needed if forChecking.
forChecking creates a file per observation and sort them in class by
class folders that the expert can then review.
Check README for more information.
"""
if self._verbatim > 0:
print('\n\n *** DATASET DISPLAY ***')
tStartGlobal = time.time()
for file_path in self.files:
recording = Recording(file_path, config, verbatim=self._verbatim)
recording.load(config)
tStart = time.time()
recording.display(config,
onlineDisplay=onlineDisplay,
saveDisplay=saveDisplay,
forChecking=forChecking,
labelEncoder=labelEncoder)
tEnd = time.time()
if self._verbatim > 1:
print('\tRecording has been loaded and displayed', file_path,
tEnd - tStart)
tEndGlobal = time.time()
if self._verbatim > 0:
print('Dataset has been displayed', tEndGlobal - tStartGlobal)
return
def load(self):
treestr = self.level*TAB + self.name + '/'
# print string to tree hierarchy and screen
self.writetree(treestr + '\n')
print(treestr)
# collect recording names: 1st char of each name must be a digit, that's all:
rnames = [ name for name in os.listdir(self.path)
if os.path.isdir(os.path.join(self.path, name))
and name[0].isdigit() ]
rnames.sort() # alphabetical order
dt = 0 # calculate total track duration by summing durations of all recordings
for rname in rnames:
path = os.path.join(self.path, rname)
recording = Recording(path, track=self)
recording.load()
self.r[recording.id] = recording
self.__setattr__('r' + str(recording.id), recording) # add shortcut attrib
dt += recording.dt
self.rnames = rnames # easy way to print out all recording names
self.dt = dt
self.dtsec = self.dt / 1e6
self.dtmin = self.dtsec / 60
self.dthour = self.dtmin / 60
# create a TrackSort with TrackNeurons:
self.sort = TrackSort(self)
self.sort.load()
# one way of calculating self.trange:
#tranges = np.asarray([ n.trange for n in self.alln.values() ])
#self.trange = min(tranges[:, 0]), max(tranges[:, 1])
# better way of calculating self.trange:
rids = sorted(self.r.keys()) # all recording ids in self
r0 = self.r[rids[0]]
r1 = self.r[rids[-1]]
assert r0.datetime == self.datetime
self.trange = r0.td+r0.trange[0], r1.td+r1.trange[1]
self.calc_meanrates()
# pttype better be the same for all member recordings:
pttype = self.r[rids[0]].pttype # init to pttype of first recording
for rid in rids[1:]:
r = self.r[rid]
# if recording doesn't have a pttype, it's probably from an old .spk file,
# so don't bother doing this test:
if hasattr(r, 'pttype') and pttype != r.pttype:
raise ValueError("inconsistent polytrode types %r and %r in track %s"
% (pttype, r.pttype, self.id))
def load(self):
treestr = self.level * TAB + self.name + '/'
# print string to tree hierarchy and screen
self.writetree(treestr + '\n')
print(treestr)
dirnames = [
name for name in os.listdir(self.path)
if os.path.isdir(os.path.join(self.path, name))
]
# collect recording names: either the 1st char of each name must be a digit,
# or the last _ separated field must be an 'e' (for 'experiment') followed by a number:
rnames = []
for dirname in dirnames:
if dirname[0].isdigit():
rnames.append(dirname)
else:
lastfield = dirname.split('_')[-1]
if lastfield[0] == 'e' and lastfield[1:].isnumeric():
rnames.append(dirname)
rnames.sort() # alphabetical order
dt = 0 # calculate total track duration by summing durations of all recordings
# does this track have any missing sorts, or rely on old impoverished .spk files?:
missingsort, simplesort = False, False
for rname in rnames:
path = os.path.join(self.path, rname)
recording = Recording(path, track=self)
recording.load()
if recording.sort == None:
missingsort = True
elif type(
recording.sort.header) in [core.SPKHeader, core.MATHeader]:
simplesort = True
self.r[recording.id] = recording
self.__setattr__('r' + str(recording.id),
recording) # add shortcut attrib
dt += recording.dt
self.rnames = rnames # easy way to print out all recording names
self.dt = dt
self.dtsec = self.dt / 1e6
self.dtmin = self.dtsec / 60
self.dthour = self.dtmin / 60
if len(rnames) == 0:
return # no recordings in this track, nothing else to do
if missingsort or simplesort:
return # skip all below due to missing or impoverished sort files (.mat or .spk)
# create a TrackSort with TrackNeurons:
self.sort = TrackSort(self)
self.sort.load()
# load RF type for each cell, should be one big dict indexed by nid:
rftypefname = os.path.join(self.path, self.absname + '.rftype')
try:
with open(rftypefname, 'r') as f:
rftypestr = f.read()
rftypes = eval(rftypestr)
for nid, rftype in rftypes.items():
assert rftype in ['simple', 'complex', 'LGN', None]
self.alln[nid].rftype = rftype
except IOError: # no absname.rftype file denoting RF type of each cell
pass
# load spike type for each cell, should be one big dict indexed by nid:
spiketypefname = os.path.join(self.path, self.absname + '.spiketype')
try:
with open(spiketypefname, 'r') as f:
spiketypestr = f.read()
spiketypes = eval(spiketypestr)
for nid, spiketype in spiketypes.items():
assert spiketype in ['fast', 'slow', 'fastasym', 'slowasym']
self.alln[nid].spiketype = spiketype
except IOError: # no absname.spiketype file denoting RF type of each cell
pass
# calculate tranges, representing start and stop times (us) of child recordings
# relative to start of track:
rids = sorted(self.r) # all recording ids in self
r0 = self.r[rids[0]]
assert r0.datetime == self.datetime
tranges = []
for rid in rids:
rec = self.r[rid]
# rec.td is time delta (us) between start of track and start of recording
trange = rec.td + rec.trange[0], rec.td + rec.trange[1]
tranges.append(trange)
self.tranges = np.array(tranges) # each row is a recording trange
self.trange = self.tranges[0, 0], self.tranges[-1, 1]
self.calc_meanrates()
# pttype better be the same for all member recordings:
pttype = self.r[rids[0]].pttype # init to pttype of first recording
for rid in rids[1:]:
r = self.r[rid]
# if recording doesn't have a pttype, it's probably from an old .spk file,
# so don't bother doing this test:
if hasattr(r, 'pttype') and pttype != r.pttype:
raise ValueError(
"inconsistent polytrode types %r and %r in track %s" %
(pttype, r.pttype, self.id))
def load(self):
treestr = self.level*TAB + self.name + '/'
# print string to tree hierarchy and screen
self.writetree(treestr + '\n')
print(treestr)
# collect recording names: 1st char of each name must be a digit, that's all:
rnames = [ name for name in os.listdir(self.path)
if os.path.isdir(os.path.join(self.path, name))
and name[0].isdigit() ]
rnames.sort() # alphabetical order
dt = 0 # calculate total track duration by summing durations of all recordings
# does this track have any missing sorts, or rely on old impoverished .spk files?:
missingsort, spksort = False, False
for rname in rnames:
path = os.path.join(self.path, rname)
recording = Recording(path, track=self)
recording.load()
if recording.sort == None:
missingsort = True
elif type(recording.sort.header) == core.SPKHeader:
spksort = True
self.r[recording.id] = recording
self.__setattr__('r' + str(recording.id), recording) # add shortcut attrib
dt += recording.dt
self.rnames = rnames # easy way to print out all recording names
self.dt = dt
self.dtsec = self.dt / 1e6
self.dtmin = self.dtsec / 60
self.dthour = self.dtmin / 60
if len(rnames) == 0:
return # no recordings in this track, nothing else to do
if missingsort or spksort:
return # skip all below due to missing .ptcs or use of impoverished .spk files
# create a TrackSort with TrackNeurons:
self.sort = TrackSort(self)
self.sort.load()
# load RF type for each cell, should be one big dict indexed by nid:
rftypefname = os.path.join(self.path, self.absname + '.rftype')
try:
with open(rftypefname, 'r') as f:
rftypestr = f.read()
rftypes = eval(rftypestr)
for nid, rftype in rftypes.items():
assert rftype in ['simple', 'complex', 'LGN', None]
self.alln[nid].rftype = rftype
except IOError: # no absname.rftype file denoting RF type of each cell
pass
# load spike type for each cell, should be one big dict indexed by nid:
spiketypefname = os.path.join(self.path, self.absname + '.spiketype')
try:
with open(spiketypefname, 'r') as f:
spiketypestr = f.read()
spiketypes = eval(spiketypestr)
for nid, spiketype in spiketypes.items():
assert spiketype in ['fast', 'slow', 'fastasym', 'slowasym']
self.alln[nid].spiketype = spiketype
except IOError: # no absname.spiketype file denoting RF type of each cell
pass
# calculate tranges, representing start and stop times (us) of child recordings
# relative to start of track:
rids = sorted(self.r.keys()) # all recording ids in self
r0 = self.r[rids[0]]
assert r0.datetime == self.datetime
tranges = []
for rid in rids:
rec = self.r[rid]
# rec.td is time delta (us) between start of track and start of recording
trange = rec.td+rec.trange[0], rec.td+rec.trange[1]
tranges.append(trange)
self.tranges = np.array(tranges) # each row is a recording trange
self.trange = self.tranges[0, 0], self.tranges[-1, 1]
self.calc_meanrates()
# pttype better be the same for all member recordings:
pttype = self.r[rids[0]].pttype # init to pttype of first recording
for rid in rids[1:]:
r = self.r[rid]
# if recording doesn't have a pttype, it's probably from an old .spk file,
# so don't bother doing this test:
if hasattr(r, 'pttype') and pttype != r.pttype:
raise ValueError("inconsistent polytrode types %r and %r in track %s"
% (pttype, r.pttype, self.id))
