def vl_look():
mpl.rcParams['font.size'] = 26
animal = 66
session = 60
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
print 'npw'
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
_, good_clusters = get_good_clusters(0)
print 'kw'
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
print 'w'
L1 = 59
L2 = 70
clusters = [[] for _ in range(L2 - L1)]
for cell, spk_i in t_cells.items():
spk_i = np.unique(spk_i)
for spk in np.nonzero((spk_i < L2) & (spk_i >= L1))[0]:
clusters[spk_i[spk] - L1].append(cell)
out = zip(clusters, vl['xs'][L1:L2], vl['ys'][L1:L2], vl['vxs'][L1:L2],
vl['vys'][L1:L2])
for tt in out:
print '%s, (%.1f,%.1f), (%.1f,%.1f)' % ((str(tt[0]), ) + tt[1:])
import pdb
pdb.set_trace()
def rate_graph():
animal = 66
session = 60 # This is August 7, 2013 run
room_shape = [[-55, 55], [-55, 55]]
tetrodes = range(1, 17)
cluster_profile = 0
bin_size = 5
_, good_clusters = get_good_clusters(cluster_profile)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in tetrodes}
tpp = 1.0 * np.mean(vl["Time"][1:] - vl["Time"][:-1])
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
label_l = vl["Task"]
# rates[cell, lbl, xbin, ybin] = firing rate
rates = get_fracs(vl["xs"], vl["ys"], label_l, room_shape, bin_size, t_cells)
rates /= tpp
plot_rates(rates, label_l, t_cells)
plt.show()
def rate_graph():
animal = 66
session = 60 # This is August 7, 2013 run
room_shape = [[-55, 55], [-55, 55]]
tetrodes = range(1, 17)
cluster_profile = 0
bin_size = 5
_, good_clusters = get_good_clusters(cluster_profile)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in tetrodes}
tpp = 1.0 * np.mean(vl['Time'][1:] - vl['Time'][:-1])
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
label_l = vl['Task']
# rates[cell, lbl, xbin, ybin] = firing rate
rates = get_fracs(vl['xs'], vl['ys'], label_l, room_shape, bin_size,
t_cells)
rates /= tpp
plot_rates(rates, label_l, t_cells)
plt.show()
def vl_look():
mpl.rcParams['font.size'] = 26
animal=66
session=60
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
print 'npw'
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
_, good_clusters = get_good_clusters(0)
print 'kw'
t_cells = count_cells(vl,cls,trigger_tm,good_clusters)
print 'w'
L1 = 59
L2 = 70
clusters = [[] for _ in range(L2-L1)]
for cell, spk_i in t_cells.items():
spk_i = np.unique(spk_i)
for spk in np.nonzero((spk_i<L2) & (spk_i>=L1))[0]:
clusters[spk_i[spk]-L1].append(cell)
out = zip(clusters,
vl['xs'][L1:L2],
vl['ys'][L1:L2],
vl['vxs'][L1:L2],
vl['vys'][L1:L2])
for tt in out:
print '%s, (%.1f,%.1f), (%.1f,%.1f)'%( (str(tt[0]),) + tt[1:])
import pdb; pdb.set_trace()
def run():
logging.basicConfig(level=logging.INFO)
good_trials = try_cache("Good trials")
animal_sess_combs = [(animal, session) for animal in [66, 70] for session in good_trials[animal]]
_, good_clusters = get_good_clusters(0)
for animal, session in animal_sess_combs:
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
for tetrode, cl in cls.items():
if tetrode not in good_clusters:
import pdb
pdb.set_trace()
continue
for cell in good_clusters[tetrode]:
logging.info(
"Finding spike locations for animal %i, session %i, tetrode %i, cell %i",
animal,
session,
tetrode,
cell,
)
cache_key = (cl["Label"][::10], vl["xs"][::10], trigger_tm, cell)
spk_i = spike_loc(cl, vl, trigger_tm, cell, key=None)
if spk_i is np.NAN:
break
store_in_cache(cache_key, spk_i)
def mk_grph(animal, session):
room_shape = [[-60, 60], [-60, 60]]
cntrx = cntry = 0
fn, _ = load_mux(animal, session)
vl = load_vl(animal, fn)
if len(np.unique(vl['Task'])) <= 1:
raise Exception('VL Task not right.')
xs = np.linspace(0, len(vl['Task']) / vl_sample_rate / 60, len(vl['Task']))
plt.figure()
angls = count_angle(vl, room_shape)
plt.plot(xs, angls, label='Angle')
scale = (np.max(angls) - np.min(angls)) / 2.0
plt.plot(xs, vl['Task'] * scale + (np.max(angls) - scale), label='Task')
# Overwrite sections with discrepencies
orient = get_orientation(vl, cntrx, cntry)
#plt.plot(orient*np.max(angls),label='Orientation')
discrep = np.sum(orient != vl['Task'])
radial = np.sum(orient == 0)
txt = ('Discrepency data: %.3f of %i \n' + 'Radial data: %.3f of %i') % (
1.0 * (discrep - radial) / (len(orient) - radial),
len(orient) - radial, 1.0 * radial / len(orient), len(orient))
txt2 = 'Filename: %s' % (fn, )
plt.autoscale(axis='x', tight=True)
plt.text(0, plt.ylim()[0], txt)
plt.text(plt.xlim()[1], plt.ylim()[0], txt2, horizontalalignment='right')
plt.ylabel('Angle (radians)')
plt.xlabel('Time (min)')
plt.legend()
plt.title('Animal:%i Session:%i Filename:%s' % (animal, session, fn))
def run():
logging.basicConfig(level=logging.INFO)
good_trials = try_cache('Good trials')
animal_sess_combs = [(animal,session) for animal in [66,70]
for session in good_trials[animal]]
_, good_clusters = get_good_clusters(0)
for animal, session in animal_sess_combs:
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
for tetrode,cl in cls.items():
if tetrode not in good_clusters:
import pdb; pdb.set_trace()
continue
for cell in good_clusters[tetrode]:
logging.info('Finding spike locations for animal %i, session %i, tetrode %i, cell %i',animal, session, tetrode,cell)
cache_key = (cl['Label'][::10],vl['xs'][::10],trigger_tm,cell)
spk_i = spike_loc(cl, vl, trigger_tm, cell, key=None)
if spk_i is np.NAN: break
store_in_cache(cache_key,spk_i)
def find_ambiguous_data():
# Final data structure will be a dictionary:
# amb[animal][session] = (#ambig, total)
# First load cache
cache = try_cache(cache_key)
if cache is not None:
amb = cache
else:
amb = {}
# Fix center
cntrx = cntry = 0
# Animal range
animals = range(65,74)
not_task_trials = []
for animal in animals:
# Add to dictionary
if animal not in amb:
amb[animal] = {}
for session in range(1,100):
if animal in amb and session in amb[animal]: #and amb[animal][session]:
logging.info('Found (Animal %i, Session %i) in cache',animal,session)
continue
try:
fn, _ = load_mux(animal, session)
except:
logging.info('Animal %i has no sessions greater than %i',animal,session-1)
break
try:
vl = load_vl(animal,fn)
except:
traceback.print_exc()
logging.info('No data found for (Animal %i, Session %i)',animal,session)
amb[animal][session] = None
not_task_trials.append([animal,session])
continue
logging.info('Checking ambiguous data for (Animal %i, Session %i)',animal,session)
orientation = get_orientation(vl,cntrx,cntry)
# Assume that orientation and task labels are matched correctly
radial = np.sum(0 == orientation)
discrepency = np.sum(vl['Task'] != orientation)
tot = len(vl['xs'])
amb[animal][session] = (radial, discrepency, tot)
# Store to cache
store_in_cache(cache_key, amb)
return amb
def rate_graph():
#mpl.rcParams['axes.titlesize'] = 18
#mpl.rcParams['axes.labelsize'] = 18
mpl.rcParams['font.size'] = 26
animal = 66
session = 60 # This is August 7, 2013 run
room_shape = [[-55,55],[-55,55]]
tetrodes = [1]
cluster_profile = 0
bin_size = 5
_, good_clusters = get_good_clusters(cluster_profile)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in tetrodes}
t_cells = count_cells(vl,cls,trigger_tm,good_clusters)
label_l = vl['Task']
# rates[cell, lbl, xbin, ybin] = firing rate
rates = get_fracs(vl['xs'], vl['ys'], label_l, room_shape, bin_size, t_cells)
for lbl in range(len(np.unique(label_l))):
plt.figure(figsize=(10,10))
x = np.concatenate([np.arange(room_shape[0][0],room_shape[0][1],bin_size),[room_shape[0][1]]])
y = np.concatenate([np.arange(room_shape[1][0],room_shape[1][1],bin_size),[room_shape[1][1]]])
Xs, Ys = np.meshgrid(x, y)
cntr = plt.pcolor(Ys,Xs,rates[3,lbl])
t=plt.colorbar(cntr, extend='both')
t.set_label('Frequency (Hz)')
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
if lbl == 0: plt.title('Clockwise')
else: plt.title('Counterclockwise')
#plt.axis('equal')
plt.xlim(room_shape[0])
plt.ylim(room_shape[1])
'''
plt.figure()
x = np.arange(room_shape[0][0],room_shape[0][1],bin_size)
y = np.arange(room_shape[1][0],room_shape[1][1],bin_size)
Xs, Ys = np.meshgrid(x, y)
cntr = plt.contourf(Ys,Xs,rate_dict[contxt][2])
t = plt.colorbar(cntr, extend='both')
t.set_label('Frequency (Hz)')
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')'''
plt.show()
def dp_accuracy():
logging.basicConfig(level=10) # 5 for more stuff
CL = CL3
animal = 66
session = 60
room =[[-55,55],[-55,55]]
bin_size = 5
K = 50 # Segment length used to calculate firing rates
CL.delt_t = K*.02
cluster_profile = 0
label = 'Task'
cl_prof_name, good_clusters = get_good_clusters(cluster_profile)
try:
adat = try_cache('One big data structure')
correct_dp = adat[CL.name][animal][session][cl_prof_name][bin_size][label][K]
logging.info('Got data from Cache.cache.')
except:
logging.info('Calculating classifications...')
CL.delt_t=K
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
label_l = vl['Task']
t_cells = count_cells(vl,cls,trigger_tm,good_clusters)
logging.info('About to generate population vector.')
#X, Y = gpv(vl, t_cells, label_l, K)
s=time()
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room)
logging.info('%i population vectors generated in %.3f.',X.shape[0],time()-s)
Y = Y.reshape([-1])
correct_dp = check_classifier(range(X.shape[0]),range(X.shape[0]),
X, Y, CL, room, bin_size)
# Accuracy meter
plt.figure()
plt.hist(correct_dp,normed=True)
plt.xlabel('Accuracy')
tt = '%s, K: %i, ClPr: %s, Label:%s'%(CL.name,K,cl_prof_name,
label)
plt.title(tt)
msg = []
for i in [1,50,75,90,95,99]:
perc = 1.0*np.sum(correct_dp > i/100.0)/len(correct_dp)*100.0
msg.append('>%i%%: %.1f%%'%(i,perc))
msg = '\n'.join(msg)
plt.xlim([0,1])
xcoord = plt.xlim()[0] + (plt.xlim()[1]-plt.xlim()[0])*.1
ycoord = plt.ylim()[0] + (plt.ylim()[1]-plt.ylim()[0])*.5
plt.text(xcoord,ycoord,msg)
plt.show()
def cluster_graphs():
animal = 70
session = 8
# Filenames (fn) are named descriptively:
# session 18:14:04 on day 10/25/2013
# load virmenLog75\20131025T181404.cmb.mat
#for tetrode in range(1,17):
for tetrode in [13]:
for context in [1, -1]:
global clrs
clrs = ['b', 'g', 'r', 'c', 'm', 'k', 'b', 'g', 'r', 'c', 'm', 'k']
fn, trigger_tm = load_mux(animal, session)
cl = load_cl(animal, fn, tetrode)
vl = load_vl(animal, fn)
spk_is = []
#for cell in range(2,100):
for cell in [3]:
cache_key = (cl['Label'][::10], vl['xs'][::10], trigger_tm,
cell)
spk_i = spike_loc(cl, vl, trigger_tm, cell, cache_key)
if spk_i is np.NAN: break
cntx_is = np.nonzero(vl['Task'] == context)[0]
spk_i = np.intersect1d(cntx_is, spk_i)
spk_is.append(spk_i)
tot_spks = len(spk_is)
if tot_spks == 0: continue
subp_x, subp_y = get_subplot_size(tot_spks)
#plt.figure()
for spk_i, i in zip(spk_is, range(tot_spks)):
plt.subplot(subp_x, subp_y, i + 1)
if context == 1:
plt.plot(vl['xs'], vl['ys'], zorder=1, color='k')
plt.scatter(vl['xs'][spk_i],
vl['ys'][spk_i],
zorder=2,
color='b',
label='Clockwise')
else:
plt.scatter(vl['xs'][spk_i],
vl['ys'][spk_i],
zorder=2,
color='r',
label='Counterclockwise')
#plot_spks(vl, spk_i, i+2)
#plt.suptitle('Animal %i, Tetrode %i, Session %i, Context:%i'%(animal,tetrode,session,context))
plt.xlim([-60, 60])
plt.ylim([-60, 60])
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.show()
def smoothing():
logging.basicConfig(level=logging.INFO)
room = [[-55, 55], [-55, 55]]
bin_size = 5
xs = range(room[0][0], room[0][1], bin_size)
ys = range(room[1][0], room[1][1], bin_size)
X, Y = np.meshgrid(xs, ys)
session = 60
animal = 66
_, good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
x = vl['xs']
y = vl['ys']
tpp = np.mean(vl['Time'][1:] - vl['Time'][:-1]) * 24 * 60 * 60
label_l = vl['Task']
# rates[cell id, lbl, xbin, ybin] = rate
rates1 = get_fracs(x,
y,
label_l,
room,
bin_size,
t_cells,
smooth_flag=True)
rates1 /= tpp
logging.info('Got smoothed rates')
rates2 = get_fracs(x,
y,
label_l,
room,
bin_size,
t_cells,
smooth_flag=False)
rates2 /= tpp
logging.info('Got unsmoothed rates')
for i in range(5): # or rates1.shape[0]
logging.info('Cell %i', i)
plt.figure()
plt.pcolor(X, Y, rates1[i, 0])
plt.colorbar()
plt.autoscale(tight=True)
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.figure()
plt.pcolor(rates2[i, 0])
plt.autoscale(tight=True)
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.show()
def cluster_graphs():
animal = 70
session = 8
# Filenames (fn) are named descriptively:
# session 18:14:04 on day 10/25/2013
# load virmenLog75\20131025T181404.cmb.mat
#for tetrode in range(1,17):
for tetrode in [13]:
for context in [1,-1]:
global clrs
clrs = ['b','g','r','c','m','k','b','g','r','c','m','k']
fn, trigger_tm = load_mux(animal, session)
cl = load_cl(animal,fn,tetrode)
vl = load_vl(animal,fn)
spk_is = []
#for cell in range(2,100):
for cell in [3]:
cache_key = (cl['Label'][::10],vl['xs'][::10],trigger_tm,cell)
spk_i = spike_loc(cl, vl, trigger_tm, cell,cache_key)
if spk_i is np.NAN: break
cntx_is = np.nonzero(vl['Task']==context)[0]
spk_i = np.intersect1d(cntx_is, spk_i)
spk_is.append(spk_i)
tot_spks = len(spk_is)
if tot_spks == 0: continue
subp_x, subp_y = get_subplot_size(tot_spks)
#plt.figure()
for spk_i, i in zip(spk_is, range(tot_spks)):
plt.subplot(subp_x,subp_y, i+1)
if context==1:
plt.plot(vl['xs'],vl['ys'],zorder=1,color='k')
plt.scatter(vl['xs'][spk_i],vl['ys'][spk_i],zorder=2,color='b',
label='Clockwise')
else:
plt.scatter(vl['xs'][spk_i],vl['ys'][spk_i],zorder=2,color='r',
label='Counterclockwise')
#plot_spks(vl, spk_i, i+2)
#plt.suptitle('Animal %i, Tetrode %i, Session %i, Context:%i'%(animal,tetrode,session,context))
plt.xlim([-60,60])
plt.ylim([-60,60])
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.show()
def smoothing():
logging.basicConfig(level=logging.INFO)
room = [[-55,55],[-55,55]]
bin_size = 5
xs = range(room[0][0],room[0][1],bin_size)
ys = range(room[1][0],room[1][1],bin_size)
X,Y = np.meshgrid(xs,ys)
session = 60
animal=66
_, good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
t_cells = count_cells(vl,cls,trigger_tm,good_clusters)
x = vl['xs']
y = vl['ys']
tpp = np.mean(vl['Time'][1:]-vl['Time'][:-1])*24*60*60
label_l = vl['Task']
# rates[cell id, lbl, xbin, ybin] = rate
rates1 = get_fracs(x,y,label_l, room, bin_size, t_cells, smooth_flag=True)
rates1 /= tpp
logging.info('Got smoothed rates')
rates2 = get_fracs(x,y,label_l, room, bin_size, t_cells, smooth_flag=False)
rates2 /= tpp
logging.info('Got unsmoothed rates')
for i in range(5): # or rates1.shape[0]
logging.info('Cell %i',i)
plt.figure()
plt.pcolor(X,Y,rates1[i,0])
plt.colorbar()
plt.autoscale(tight=True)
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.figure()
plt.pcolor(rates2[i,0])
plt.autoscale(tight=True)
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
plt.show()
def gpv_rates():
logging.basicConfig(level=logging.INFO)
animal = 66
session = 60
room_shape = [[-55, 55], [-55, 55]]
tetrodes = [1]
cells = range(2, 10)
bin_size = 5
K = 50 # Segment length used to calculate firing rates
#xbins = ybins = (room_shape[0][1]-room_shape[0][0])/bin_size
good_clusters = {tetrode: cells for tetrode in tetrodes}
#good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in tetrodes}
label_l = vl['Task']
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room_shape)
logging.info('%i population vectors generated.', X.shape[0])
Y = Y.reshape([-1])
# GPV rates
rates = fracs_from_pv(X, Y, bin_size, room_shape, smooth_flag=True)
# Now get normally calculate rates
real_rates = get_fracs(vl['xs'],
vl['ys'],
label_l,
room_shape,
bin_size,
t_cells,
smooth_flag=True)
try:
assert np.all(rates == real_rates)
except:
print 'DOESNT WORK!!'
plot_rates(rates, Y, t_cells)
plot_rates(real_rates, Y, t_cells)
plt.show()
def count():
logging.basicConfig(level=logging.INFO)
animals = [66,70]
sessions = range(1,100)
tetrodes = range(1,17)
good_trials = try_cache('Good trials')
key = (animals,sessions,tetrodes,'count clusters')
cache = try_cache(key)
cache=None
if cache is not None:
cls = cache
else:
cls = []
for animal in animals:
print 'Animal ', animal
for session in good_trials[animal]:
print 'Session', session
fn, _ = load_mux(animal, session)
vl = load_vl(animal,fn)
if len(np.unique(vl['Task'])) != 2:continue
cells = 0
for tetrode in tetrodes:
cl = load_cl(animal, fn, tetrode)
cells += len(np.unique(cl['Label']))-1
if cells == 0: continue
cls.append((animal, session, cells,len(vl)))
store_in_cache(key,cls)
cls.sort(key=lambda x:x[2])
txt = '%i %i %i %i'
print 'Animal Session Cells Length'
for animal,session,cells, length in cls:
print txt%(animal,session,cells, length)
import pdb; pdb.set_trace()
print 'Mean length:', np.mean([cl[3] for cl in cls])
def run():
logging.basicConfig(level=logging.INFO)
cache_key = 'Good trials'
animals = [66, 73]
sessions = range(100)
_, good_clusters = goodClusters.get_good_clusters(0)
good_trials = try_cache(cache_key)
if good_trials is None: good_trials = {}
for animal in animals:
if animal not in good_trials: good_trials[animal] = []
for session in sessions:
if session in good_trials[animal]: continue
try:
fn, trigger_tm = load_mux(animal, session)
except:
logging.info('Animal %i has no sessions greater than %i',
animal, session + 1)
break
try:
vl = load_vl(animal, fn)
except:
logging.info('Animal %i session %i is not a task trial',
animal, session + 1)
continue
cls = {
tetrode: load_cl(animal, fn, tetrode)
for tetrode in range(1, 17)
}
try:
count_cells(vl, cls, trigger_tm, good_clusters)
except:
# No cells found
continue
if session not in good_trials[animal]:
good_trials[animal].append(session)
store_in_cache(cache_key, good_trials)
def count():
logging.basicConfig(level=logging.INFO)
animals = [66, 70]
sessions = range(1, 100)
tetrodes = range(1, 17)
good_trials = try_cache('Good trials')
key = (animals, sessions, tetrodes, 'count clusters')
cache = try_cache(key)
cache = None
if cache is not None:
cls = cache
else:
cls = []
for animal in animals:
print 'Animal ', animal
for session in good_trials[animal]:
print 'Session', session
fn, _ = load_mux(animal, session)
vl = load_vl(animal, fn)
if len(np.unique(vl['Task'])) != 2: continue
cells = 0
for tetrode in tetrodes:
cl = load_cl(animal, fn, tetrode)
cells += len(np.unique(cl['Label'])) - 1
if cells == 0: continue
cls.append((animal, session, cells, len(vl)))
store_in_cache(key, cls)
cls.sort(key=lambda x: x[2])
txt = '%i %i %i %i'
print 'Animal Session Cells Length'
for animal, session, cells, length in cls:
print txt % (animal, session, cells, length)
import pdb
pdb.set_trace()
print 'Mean length:', np.mean([cl[3] for cl in cls])
def gpv_rates():
logging.basicConfig(level=logging.INFO)
animal = 66
session = 60
room_shape = [[-55,55],[-55,55]]
tetrodes = [1]
cells = range(2,10)
bin_size = 5
K = 50# Segment length used to calculate firing rates
#xbins = ybins = (room_shape[0][1]-room_shape[0][0])/bin_size
good_clusters = {tetrode:cells for tetrode in tetrodes}
#good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in tetrodes}
label_l = vl['Task']
t_cells = count_cells(vl,cls,trigger_tm, good_clusters)
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room_shape)
logging.info('%i population vectors generated.',X.shape[0])
Y = Y.reshape([-1])
# GPV rates
rates = fracs_from_pv(X,Y,bin_size,room_shape,smooth_flag=True)
# Now get normally calculate rates
real_rates = get_fracs(vl['xs'],vl['ys'],label_l,room_shape,bin_size, t_cells,smooth_flag=True)
try:
assert np.all(rates == real_rates)
except:
print 'DOESNT WORK!!'
plot_rates(rates,Y,t_cells)
plot_rates(real_rates,Y,t_cells)
plt.show()
def count_vl():
logging.basicConfig(level=logging.INFO)
llist = []
good_trials = try_cache('Good trials').items()
for animal, sessions in good_trials:
print 'Animal', animal
for session in sessions:
print 'Session', session
fn, _ = load_mux(animal, session)
vl = load_vl(animal,fn)
llist.append(len(vl['xs']))
print np.mean(llist)
print len(llist)
plt.hist(llist)
plt.xlabel('Number of recorded points')
plt.ylabel('Count')
plt.title('Recorded Points per Session')
plt.show()
import pdb; pdb.set_trace()
def view_PCA():
animal = 66
session = 60
bin_size = 5
K = 50 # Segment length used to calculate firing rates
label = 'Task'
room = [[-55, 55], [-55, 55]]
_, good_clusters = get_good_clusters(0)
xbins = (room[0][1] - [0][0]) / bin_size
ybins = (room[1][1] - [1][0]) / bin_size
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
if label == 'Task':
label_l = vl['Task']
else:
raise Exception('Not implemented yet.')
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
logging.info('About to generate population vector.')
#X, Y = gpv(vl, t_cells, label_l, K)
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room)
pcas = np.zeros([xbins, ybins])
for xbin, ybin in product(xbins, ybins):
pca = PCA()
Xtmp = np.zeros([
X.shape[0],
])
X = pca.fit_transform(X[:, :len(t_cells)])
pcas[xbin, ybin] = pca
plt.plot(pca.explained_variance_ratio_)
print pca.components_
plt.show()
def mk_grph(animal,session):
room_shape = [[-60,60],[-60,60]]
cntrx = cntry = 0
fn, _ = load_mux(animal, session)
vl = load_vl(animal,fn)
if len(np.unique(vl['Task'])) <= 1:
raise Exception('VL Task not right.')
xs = np.linspace(0,len(vl['Task'])/vl_sample_rate/60,len(vl['Task']))
plt.figure()
angls = count_angle(vl,room_shape)
plt.plot(xs,angls,label='Angle')
scale = (np.max(angls)-np.min(angls))/2.0
plt.plot(xs,vl['Task']*scale+(np.max(angls)-scale),label='Task')
# Overwrite sections with discrepencies
orient = get_orientation(vl,cntrx,cntry)
#plt.plot(orient*np.max(angls),label='Orientation')
discrep = np.sum(orient != vl['Task'])
radial = np.sum(orient == 0)
txt = ('Discrepency data: %.3f of %i \n'+
'Radial data: %.3f of %i')%(1.0*(discrep-radial)/(len(orient)-radial),
len(orient)-radial,
1.0*radial/len(orient), len(orient))
txt2 = 'Filename: %s'%(fn,)
plt.autoscale(axis='x',tight=True)
plt.text(0,plt.ylim()[0],txt)
plt.text(plt.xlim()[1],plt.ylim()[0],txt2,horizontalalignment='right')
plt.ylabel('Angle (radians)')
plt.xlabel('Time (min)')
plt.legend()
plt.title('Animal:%i Session:%i Filename:%s'%(animal,session, fn))
def run():
logging.basicConfig(level=logging.INFO)
cache_key = 'Good trials'
animals = [66,73]
sessions = range(100)
_, good_clusters = goodClusters.get_good_clusters(0)
good_trials = try_cache(cache_key)
if good_trials is None: good_trials = {}
for animal in animals:
if animal not in good_trials: good_trials[animal] = []
for session in sessions:
if session in good_trials[animal]: continue
try:
fn, trigger_tm = load_mux(animal, session)
except:
logging.info('Animal %i has no sessions greater than %i',animal,session+1)
break
try:
vl = load_vl(animal,fn)
except:
logging.info('Animal %i session %i is not a task trial',animal,session+1)
continue
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
try:
count_cells(vl,cls,trigger_tm, good_clusters)
except:
# No cells found
continue
if session not in good_trials[animal]:
good_trials[animal].append(session)
store_in_cache(cache_key,good_trials)
def view_PCA():
animal = 66
session = 60
bin_size = 5
K = 50 # Segment length used to calculate firing rates
label = "Task"
room = [[-55, 55], [-55, 55]]
_, good_clusters = get_good_clusters(0)
xbins = (room[0][1] - [0][0]) / bin_size
ybins = (room[1][1] - [1][0]) / bin_size
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
if label == "Task":
label_l = vl["Task"]
else:
raise Exception("Not implemented yet.")
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
logging.info("About to generate population vector.")
# X, Y = gpv(vl, t_cells, label_l, K)
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room)
pcas = np.zeros([xbins, ybins])
for xbin, ybin in product(xbins, ybins):
pca = PCA()
Xtmp = np.zeros([X.shape[0]])
X = pca.fit_transform(X[:, : len(t_cells)])
pcas[xbin, ybin] = pca
plt.plot(pca.explained_variance_ratio_)
print pca.components_
plt.show()
def view_simulation():
logging.basicConfig(level=logging.INFO)
mpl.rcParams['font.size'] = 26
lw = 3
CLs = [CL6,CL7, CL2]
Folds = 6
# Good trials is a dictionary
# good_trials[animal] = [list of sessions that are task trials
# and have at least one labeled cluster]
#good_trials = try_cache('Good trials')
#animal_sess_combs = [(animal,session) for animal in range(65,74)
# for session in good_trials[animal]]
animal_sess_combs = [(66,60)]
bin_sizes = [5]
cl_profs = [0]
label = 'Task'
exceptions = []
room = [[-55,55],[-55,55]]
adat = try_cache('One big data structure for %i folds'%(Folds,))
#adat = try_cache('One big data structure')
if adat is None: raise Exception()
print adat.keys()
good_trials = try_cache('Good trials')
for animal, session in animal_sess_combs:
# Get time per point
fn, _ = load_mux(animal, session)
vl = load_vl(animal,fn)
tms = vl['Time']*24*60*60
tpp = np.mean(tms[1:]-tms[:-1])
print tpp
plt.figure()
for CL, cluster_profile, bin_size in product(CLs, cl_profs, bin_sizes):
if (CL,cluster_profile) in exceptions: continue
if animal not in adat[CL.name] or session not in adat[CL.name][animal]:continue
cl_prof_name, _ = get_good_clusters(cluster_profile)
pts = []
try:
for K, correct_dp in adat[CL.name][animal][session][cl_prof_name][bin_size][label].items():
if len(correct_dp) == 0: continue
y = 100.0*np.sum(correct_dp>.5)/len(correct_dp)
pts.append((K,y))
except:
logging.warning('Something fishy with %s',CL.name)
if len(pts) == 0: continue
pts.sort(key=lambda x: x[0])
# Get the right color
CL_i = CLs.index(CL)
cp_i = cl_profs.index(cluster_profile)
b_i = bin_sizes.index(bin_size)
clr_i = CL_i*len(cl_profs)+cp_i
clr_str = clrs[clr_i]+ln_typs[b_i]
xs,ys = zip(*pts)
plt.plot(np.array(xs)*tpp,ys,clr_str,label=CL.name,
linewidth=lw)
plt.legend(fontsize='x-small',loc='lower right')
plt.xlabel('Segment Length (s)')
plt.ylabel('Percent Correct')
#plt.title('Accuracy vs Segment Size, Animal %i Session %i'%(animal, session))
plt.ylim([60,95])
plt.title('%i-Fold Validation'%(Folds,))
plt.show()
def checkGPV():
logging.basicConfig(level=5)
animal = 66
session = 60
room_shape = [[-55, 55], [-55, 55]]
tetrodes = [1]
cells = range(2, 10)
bin_size = 5
K = 1 # Segment length used to calculate firing rates
maxs = 10000
assert maxs % K == 0
#xbins = ybins = (room_shape[0][1]-room_shape[0][0])/bin_size
good_clusters = {tetrode: cells for tetrode in tetrodes}
#good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in tetrodes}
label_l = vl['Task']
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
''''''
# For debugging purposes, make sure it's not too big
if len(label_l) > maxs:
vl['xs'] = vl['xs'][:maxs * 10:10]
vl['ys'] = vl['ys'][:maxs * 10:10]
label_l = label_l[:maxs * 10:10]
for key in t_cells:
tmp = np.array(t_cells[key])
tmp = tmp[tmp < maxs]
t_cells[key] = tmp
for gpv in [gpv1]:
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room_shape)
logging.info('%i population vectors generated.', X.shape[0])
Y = Y.reshape([-1])
# Debug code
# This is only true if no points are thrown away
if X.shape[0] * K == maxs:
tot_spks = np.sum(
[len(np.unique(spki)) for spki in t_cells.values()])
assert tot_spks == np.sum(X[:, :len(t_cells)]) * K
# GPV rates
rates = fracs_from_pv(X, Y, bin_size, room_shape, smooth_flag=False)
# Now get normally calculate rates
real_rates = get_fracs(vl['xs'],
vl['ys'],
label_l,
room_shape,
bin_size,
t_cells,
smooth_flag=False)
try:
assert np.all(rates == real_rates)
except:
print 'DOESNT WORK!!'
plot_rates(rates, Y, t_cells)
plot_rates(real_rates, Y, t_cells)
plt.show()
# sgn are the indices of task_i where
# task_i[j] != task_i[j+1]
sgn = np.nonzero(task_ip1 != task_i)[0]
sgn_i = np.concatenate([[sgn[0]],sgn])
sgn_ip1 = np.concatenate([sgn,[len(task)]])
run_len = (sgn_ip1-sgn_i)[1:]
return sgn, run_len
if __name__ == '__main__':
from matplotlib import pyplot as plt
from Data.readData import load_mux, load_vl
num = 66
session = 60
fn, _= load_mux(num,session)
vl = load_vl(num,fn)
task = get_orientation(vl,0,0)
sgn, run_len = find_runs(task)
n,bins, _ = plt.hist(run_len,bins=range(1,np.max(run_len)+1))
plt.title('Run length')
import pdb; pdb.set_trace()
plt.figure()
plt.hist(run_len,bins=range(1,np.max(run_len)+1),cumulative=-1)
plt.title('Reverse Cumulative run length')
plt.show()
def rate_graph():
#mpl.rcParams['axes.titlesize'] = 18
#mpl.rcParams['axes.labelsize'] = 18
mpl.rcParams['font.size'] = 26
animal = 66
session = 60 # This is August 7, 2013 run
room_shape = [[-55, 55], [-55, 55]]
tetrodes = [1]
cluster_profile = 0
bin_size = 5
_, good_clusters = get_good_clusters(cluster_profile)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in tetrodes}
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
label_l = vl['Task']
# rates[cell, lbl, xbin, ybin] = firing rate
rates = get_fracs(vl['xs'], vl['ys'], label_l, room_shape, bin_size,
t_cells)
for lbl in range(len(np.unique(label_l))):
plt.figure(figsize=(10, 10))
x = np.concatenate([
np.arange(room_shape[0][0], room_shape[0][1], bin_size),
[room_shape[0][1]]
])
y = np.concatenate([
np.arange(room_shape[1][0], room_shape[1][1], bin_size),
[room_shape[1][1]]
])
Xs, Ys = np.meshgrid(x, y)
cntr = plt.pcolor(Ys, Xs, rates[3, lbl])
t = plt.colorbar(cntr, extend='both')
t.set_label('Frequency (Hz)')
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')
if lbl == 0: plt.title('Clockwise')
else: plt.title('Counterclockwise')
#plt.axis('equal')
plt.xlim(room_shape[0])
plt.ylim(room_shape[1])
'''
plt.figure()
x = np.arange(room_shape[0][0],room_shape[0][1],bin_size)
y = np.arange(room_shape[1][0],room_shape[1][1],bin_size)
Xs, Ys = np.meshgrid(x, y)
cntr = plt.contourf(Ys,Xs,rate_dict[contxt][2])
t = plt.colorbar(cntr, extend='both')
t.set_label('Frequency (Hz)')
plt.xlabel('Position (in)')
plt.ylabel('Position (in)')'''
plt.show()
def run(Folds):
# Toggle-able parameters
#CLs = [CL2,CL6,CL5]
#CLs = [CL6, CL7]
CLs = [CL10]
Ks = np.arange(10,200,20) # Segment length used to calculate firing rates
# Sort of toggle-able parameters
#animal_sess_combs = [(66,60),(70,8),(70,10),(66,61)]
animal_sess_combs = [(66,60)]
#good_trials = try_cache('Good trials')
#animal_sess_combs = [(animal,session) for animal in range(65,74)
# for session in good_trials[animal]]
bin_sizes = [5]
label = 'Task'
exceptions = []
cl_profs = [0]
# Not really toggle-able parameters
room = [[-55,55],[-55,55]]
cache = try_cache('One big data structure for %i folds'%(Folds,))
adat = ({} if cache is None else cache)
for animal, session in animal_sess_combs:
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
if label == 'Task': label_l = vl['Task']
else: raise Exception('Not implemented yet.')
for clust_prof in cl_profs:
cl_prof_name, good_clusters = get_good_clusters(clust_prof)
t_cells = count_cells(vl,cls,trigger_tm,good_clusters)
for bin_size, K in product(bin_sizes,Ks):
cached = np.zeros(len(CLs))
for CL in CLs:
i = CLs.index(CL)
try:
raise Exception
adat[CL.name][animal][session][cl_prof_name][bin_size][label][K]
cached[i] = True
except:
cached[i] = False
if np.sum(cached) == len(CLs):
print 'Everything already cached'
continue # Everything is already cached!
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room)
# The main data stricture
dps = {CL:[] for CL in CLs if CL not in cached}
if Folds >0: kf = cross_validation.KFold(len(Y),n_folds=Folds,shuffle=True)
else: kf = [(range(len(Y)),range(len(Y)))]
for train_index, test_index in kf:
logging.warning('Training/testing: %i/%i',len(train_index),len(test_index))
for CL in CLs:
if cached[CLs.index(CL)]: continue
logging.warning('%s, %i seg, (%i, %i)',CL.name, K, animal, session)
if (CL,clust_prof) in exceptions: continue
CL.delt_t = K
correct_dp = check_classifier(train_index,test_index,X,Y,CL, room, bin_size)
dps[CL].extend(correct_dp.tolist())
for CL in CLs:
if cached[CLs.index(CL)]: continue
to_add = np.array(dps[CL]).reshape([-1])
add(adat, CL.name, animal, session, cl_prof_name, bin_size, label, K, to_add)
store_in_cache('One big data structure for %i folds'%(Folds,),adat)
def run(Folds):
# Toggle-able parameters
#CLs = [CL2,CL6,CL5]
#CLs = [CL6, CL7]
CLs = [CL10]
Ks = np.arange(10, 200,
20) # Segment length used to calculate firing rates
# Sort of toggle-able parameters
#animal_sess_combs = [(66,60),(70,8),(70,10),(66,61)]
animal_sess_combs = [(66, 60)]
#good_trials = try_cache('Good trials')
#animal_sess_combs = [(animal,session) for animal in range(65,74)
# for session in good_trials[animal]]
bin_sizes = [5]
label = 'Task'
exceptions = []
cl_profs = [0]
# Not really toggle-able parameters
room = [[-55, 55], [-55, 55]]
cache = try_cache('One big data structure for %i folds' % (Folds, ))
adat = ({} if cache is None else cache)
for animal, session in animal_sess_combs:
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {
tetrode: load_cl(animal, fn, tetrode)
for tetrode in range(1, 17)
}
if label == 'Task': label_l = vl['Task']
else: raise Exception('Not implemented yet.')
for clust_prof in cl_profs:
cl_prof_name, good_clusters = get_good_clusters(clust_prof)
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
for bin_size, K in product(bin_sizes, Ks):
cached = np.zeros(len(CLs))
for CL in CLs:
i = CLs.index(CL)
try:
raise Exception
adat[CL.name][animal][session][cl_prof_name][bin_size][
label][K]
cached[i] = True
except:
cached[i] = False
if np.sum(cached) == len(CLs):
print 'Everything already cached'
continue # Everything is already cached!
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room)
# The main data stricture
dps = {CL: [] for CL in CLs if CL not in cached}
if Folds > 0:
kf = cross_validation.KFold(len(Y),
n_folds=Folds,
shuffle=True)
else:
kf = [(range(len(Y)), range(len(Y)))]
for train_index, test_index in kf:
logging.warning('Training/testing: %i/%i',
len(train_index), len(test_index))
for CL in CLs:
if cached[CLs.index(CL)]: continue
logging.warning('%s, %i seg, (%i, %i)', CL.name, K,
animal, session)
if (CL, clust_prof) in exceptions: continue
CL.delt_t = K
correct_dp = check_classifier(train_index, test_index,
X, Y, CL, room, bin_size)
dps[CL].extend(correct_dp.tolist())
for CL in CLs:
if cached[CLs.index(CL)]: continue
to_add = np.array(dps[CL]).reshape([-1])
add(adat, CL.name, animal, session, cl_prof_name, bin_size,
label, K, to_add)
store_in_cache('One big data structure for %i folds' % (Folds, ), adat)
def generate_DP_accuracy_graph():
animal = 66
session = 60 # This is August 7, 2013 run
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in range(1,17)}
room_shape = [[-60,60],[-60,60]]
bin_size = 8
# Actual Contexts
'''
cached = try_cache(Classifier,Classifier.name,vl,cls,trigger_tm,label_is,room_shape,bin_size)
if cached is not None:
classifier, Xs, Ys = cached
logging.info('Got classifier and population vectors from Cache.cache.')
else:
classifier = Classifier(vl,cls,trigger_tm, label_is, room_shape, bin_size)
Xs, Ys = classifier.generate_population_vectors()
store_in_cache(Classifier,Classifier.name,vl,cls,trigger_tm,label_is,room_shape,bin_size,
[classifier,Xs,Ys])'''
# Label based on task
'''
labels = np.unique(vl['Task'])
label_is = {contxt: np.nonzero(vl['Task']==contxt)[0] for contxt in labels}
label_l = vl['Task']'''
''''''
label_l = get_orientation(vl,cntrx=0,cntry=0)
labels = np.unique(label_l)
label_is = {contxt: np.nonzero(label_l==contxt)[0] for contxt in labels}
classifier = Classifier(vl,cls,trigger_tm, label_is, room_shape, bin_size)
Xs, Ys = classifier.generate_population_vectors(label_l)
correct_dp = []
incorrect_dp = []
for (xbin,ybin),vecs,lbls in zip(Xs.keys(),Xs.values(),Ys.values()):
for vec,lbl in zip(vecs,lbls):
if lbl == 0:
crct, incrct = classifier.classifiy(xbin, ybin, vec)
else:
incrct, crct = classifier.classifiy(xbin, ybin, vec)
correct_dp.append(crct)
incorrect_dp.append(incrct)
# Process
correct_dp = np.array(correct_dp)
incorrect_dp = np.array(incorrect_dp)
nonzero_is = (correct_dp > 0) | (incorrect_dp > 0)
correct_dp = correct_dp[np.nonzero(nonzero_is)[0]]
incorrect_dp = incorrect_dp[np.nonzero(nonzero_is)[0]]
from matplotlib import pyplot as plt
# 2d Histogram
plt.figure()
hist,xedges,yedges = np.histogram2d(correct_dp, incorrect_dp, 150)
Xs, Ys = np.meshgrid(xedges, yedges)
grph = plt.pcolor(Xs,Ys,hist)
plt.xlim([0,xedges[-1]])
plt.ylim([0,yedges[-1]])
plt.colorbar(grph, extend='both')
plt.title('Dot Product Classifier Accuracy')
plt.xlabel('Population vector x Correct Template')
plt.ylabel('Population vector x Incorrect Template')
# Accuracy meter
plt.figure()
accuracy = correct_dp / np.sqrt(correct_dp**2+incorrect_dp**2)
plt.hist(accuracy,normed=True)
plt.xlabel('Accuracy')
plt.title(classifier.name)
msg = []
for i in [1,50,75,90,95,99]:
perc = 1.0*np.sum(accuracy > i/100.0)/len(accuracy)*100.0
msg.append('>%i%%: %.1f%%'%(i,perc))
msg = '\n'.join(msg)
plt.xlim([0,1])
xcoord = plt.xlim()[0] + (plt.xlim()[1]-plt.xlim()[0])*.1
ycoord = plt.ylim()[0] + (plt.ylim()[1]-plt.ylim()[0])*.5
plt.text(xcoord,ycoord,msg)
plt.show()
#from ContextPredictors.DotProduct import DotProduct
timeout_rate = 15
animation_step_size = 10
if __name__ == '__main__':
logging.basicConfig(level=logging.DEBUG)
room_shape = [[-55,55],[-55,55]]
animal = 66
session = 60
tetrode=3
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cl = load_cl(animal,fn,tetrode)
wv = load_wv(animal,fn,tetrode)
wv_iters = match_cl_to_vl(cl['Time'],vl, trigger_tm)
#WAVE_L = WR.read(vl[2],vl[3], is_clockwise(*vl[2:6]))
#actual_prediction = np.array([is_clockwise(x,y,vx,vy) for
# x,y,vx,vy in zip(*vl.values()[2:6])])
#logging.info('Starting to generate waveforms...')
#global WAVE_L
#WAVE_L = WR.read(vl[2],vl[3],actual_prediction)
#WAVE_L = np.array([WR.read(x,y,actual) for actual, x,y in zip(actual_prediction, *vl[2:4])])
#logging.info('Finished generating waveform.')
'''
HMM = PiecewiseHMM(vl[2],vl[3],WAVE_L,actual_prediction)
def find_ambiguous_data():
# Final data structure will be a dictionary:
# amb[animal][session] = (#ambig, total)
# First load cache
cache = try_cache(cache_key)
if cache is not None:
amb = cache
else:
amb = {}
# Fix center
cntrx = cntry = 0
# Animal range
animals = range(65, 74)
not_task_trials = []
for animal in animals:
# Add to dictionary
if animal not in amb:
amb[animal] = {}
for session in range(1, 100):
if animal in amb and session in amb[
animal]: #and amb[animal][session]:
logging.info('Found (Animal %i, Session %i) in cache', animal,
session)
continue
try:
fn, _ = load_mux(animal, session)
except:
logging.info('Animal %i has no sessions greater than %i',
animal, session - 1)
break
try:
vl = load_vl(animal, fn)
except:
traceback.print_exc()
logging.info('No data found for (Animal %i, Session %i)',
animal, session)
amb[animal][session] = None
not_task_trials.append([animal, session])
continue
logging.info('Checking ambiguous data for (Animal %i, Session %i)',
animal, session)
orientation = get_orientation(vl, cntrx, cntry)
# Assume that orientation and task labels are matched correctly
radial = np.sum(0 == orientation)
discrepency = np.sum(vl['Task'] != orientation)
tot = len(vl['xs'])
amb[animal][session] = (radial, discrepency, tot)
# Store to cache
store_in_cache(cache_key, amb)
return amb
sgn = np.nonzero(task_ip1 != task_i)[0]
sgn_i = np.concatenate([[sgn[0]], sgn])
sgn_ip1 = np.concatenate([sgn, [len(task)]])
run_len = (sgn_ip1 - sgn_i)[1:]
return sgn, run_len
if __name__ == '__main__':
from matplotlib import pyplot as plt
from Data.readData import load_mux, load_vl
num = 66
session = 60
fn, _ = load_mux(num, session)
vl = load_vl(num, fn)
task = get_orientation(vl, 0, 0)
sgn, run_len = find_runs(task)
n, bins, _ = plt.hist(run_len, bins=range(1, np.max(run_len) + 1))
plt.title('Run length')
import pdb
pdb.set_trace()
plt.figure()
plt.hist(run_len, bins=range(1, np.max(run_len) + 1), cumulative=-1)
plt.title('Reverse Cumulative run length')
plt.show()
def checkGPV():
logging.basicConfig(level=5)
animal = 66
session = 60
room_shape = [[-55,55],[-55,55]]
tetrodes = [1]
cells = range(2,10)
bin_size = 5
K = 1# Segment length used to calculate firing rates
maxs = 10000
assert maxs%K==0
#xbins = ybins = (room_shape[0][1]-room_shape[0][0])/bin_size
good_clusters = {tetrode:cells for tetrode in tetrodes}
#good_clusters = get_good_clusters(0)
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal,fn)
cls = {tetrode:load_cl(animal,fn,tetrode) for tetrode in tetrodes}
label_l = vl['Task']
t_cells = count_cells(vl,cls,trigger_tm, good_clusters)
''''''
# For debugging purposes, make sure it's not too big
if len(label_l) > maxs:
vl['xs'] = vl['xs'][:maxs*10:10]
vl['ys'] = vl['ys'][:maxs*10:10]
label_l = label_l[:maxs*10:10]
for key in t_cells:
tmp = np.array(t_cells[key])
tmp = tmp[tmp<maxs]
t_cells[key] = tmp
for gpv in [gpv1]:
logging.info('About to generate population vector.')
X, Y = gpv(vl, t_cells, label_l, K, bin_size, room_shape)
logging.info('%i population vectors generated.',X.shape[0])
Y = Y.reshape([-1])
# Debug code
# This is only true if no points are thrown away
if X.shape[0]*K == maxs:
tot_spks = np.sum([len(np.unique(spki)) for spki in t_cells.values()])
assert tot_spks == np.sum(X[:,:len(t_cells)])*K
# GPV rates
rates = fracs_from_pv(X,Y,bin_size,room_shape,smooth_flag=False)
# Now get normally calculate rates
real_rates = get_fracs(vl['xs'],vl['ys'],label_l,room_shape,bin_size, t_cells,smooth_flag=False)
try:
assert np.all(rates == real_rates)
except:
print 'DOESNT WORK!!'
plot_rates(rates,Y,t_cells)
plot_rates(real_rates,Y,t_cells)
plt.show()
def generate_DP_accuracy_graph():
animal = 66
session = 60 # This is August 7, 2013 run
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
room_shape = [[-60, 60], [-60, 60]]
bin_size = 8
# Actual Contexts
'''
cached = try_cache(Classifier,Classifier.name,vl,cls,trigger_tm,label_is,room_shape,bin_size)
if cached is not None:
classifier, Xs, Ys = cached
logging.info('Got classifier and population vectors from Cache.cache.')
else:
classifier = Classifier(vl,cls,trigger_tm, label_is, room_shape, bin_size)
Xs, Ys = classifier.generate_population_vectors()
store_in_cache(Classifier,Classifier.name,vl,cls,trigger_tm,label_is,room_shape,bin_size,
[classifier,Xs,Ys])'''
# Label based on task
'''
labels = np.unique(vl['Task'])
label_is = {contxt: np.nonzero(vl['Task']==contxt)[0] for contxt in labels}
label_l = vl['Task']'''
''''''
label_l = get_orientation(vl, cntrx=0, cntry=0)
labels = np.unique(label_l)
label_is = {contxt: np.nonzero(label_l == contxt)[0] for contxt in labels}
classifier = Classifier(vl, cls, trigger_tm, label_is, room_shape,
bin_size)
Xs, Ys = classifier.generate_population_vectors(label_l)
correct_dp = []
incorrect_dp = []
for (xbin, ybin), vecs, lbls in zip(Xs.keys(), Xs.values(), Ys.values()):
for vec, lbl in zip(vecs, lbls):
if lbl == 0:
crct, incrct = classifier.classifiy(xbin, ybin, vec)
else:
incrct, crct = classifier.classifiy(xbin, ybin, vec)
correct_dp.append(crct)
incorrect_dp.append(incrct)
# Process
correct_dp = np.array(correct_dp)
incorrect_dp = np.array(incorrect_dp)
nonzero_is = (correct_dp > 0) | (incorrect_dp > 0)
correct_dp = correct_dp[np.nonzero(nonzero_is)[0]]
incorrect_dp = incorrect_dp[np.nonzero(nonzero_is)[0]]
from matplotlib import pyplot as plt
# 2d Histogram
plt.figure()
hist, xedges, yedges = np.histogram2d(correct_dp, incorrect_dp, 150)
Xs, Ys = np.meshgrid(xedges, yedges)
grph = plt.pcolor(Xs, Ys, hist)
plt.xlim([0, xedges[-1]])
plt.ylim([0, yedges[-1]])
plt.colorbar(grph, extend='both')
plt.title('Dot Product Classifier Accuracy')
plt.xlabel('Population vector x Correct Template')
plt.ylabel('Population vector x Incorrect Template')
# Accuracy meter
plt.figure()
accuracy = correct_dp / np.sqrt(correct_dp**2 + incorrect_dp**2)
plt.hist(accuracy, normed=True)
plt.xlabel('Accuracy')
plt.title(classifier.name)
msg = []
for i in [1, 50, 75, 90, 95, 99]:
perc = 1.0 * np.sum(accuracy > i / 100.0) / len(accuracy) * 100.0
msg.append('>%i%%: %.1f%%' % (i, perc))
msg = '\n'.join(msg)
plt.xlim([0, 1])
xcoord = plt.xlim()[0] + (plt.xlim()[1] - plt.xlim()[0]) * .1
ycoord = plt.ylim()[0] + (plt.ylim()[1] - plt.ylim()[0]) * .5
plt.text(xcoord, ycoord, msg)
plt.show()
t_cells[(tetrode, cell)] = spk_i
return t_cells
if __name__ == '__main__':
logging.basicConfig(level=logging.DEBUG)
animal = 66
session = 60
room_shape = [[-60, 60], [-60, 60]]
bin_size = 8
K = 32 # Segment length used to calculate firing rates
fn, trigger_tm = load_mux(animal, session)
vl = load_vl(animal, fn)
cls = {tetrode: load_cl(animal, fn, tetrode) for tetrode in range(1, 17)}
''''''
# Label with task
labels = np.unique(vl['Task'])
label_is = {
contxt: np.nonzero(vl['Task'] == contxt)[0]
for contxt in labels
}
label_l = vl['Task']
t_cells = count_cells(vl, cls, trigger_tm, good_clusters)
X, Y = gpv(vl, t_cells, room_shape, bin_size, label_l, K=32)
classifier = Classifier(X, Y)
def view_simulation():
logging.basicConfig(level=logging.INFO)
mpl.rcParams['font.size'] = 26
lw = 3
CLs = [CL6, CL7, CL2]
Folds = 6
# Good trials is a dictionary
# good_trials[animal] = [list of sessions that are task trials
# and have at least one labeled cluster]
#good_trials = try_cache('Good trials')
#animal_sess_combs = [(animal,session) for animal in range(65,74)
# for session in good_trials[animal]]
animal_sess_combs = [(66, 60)]
bin_sizes = [5]
cl_profs = [0]
label = 'Task'
exceptions = []
room = [[-55, 55], [-55, 55]]
adat = try_cache('One big data structure for %i folds' % (Folds, ))
#adat = try_cache('One big data structure')
if adat is None: raise Exception()
print adat.keys()
good_trials = try_cache('Good trials')
for animal, session in animal_sess_combs:
# Get time per point
fn, _ = load_mux(animal, session)
vl = load_vl(animal, fn)
tms = vl['Time'] * 24 * 60 * 60
tpp = np.mean(tms[1:] - tms[:-1])
print tpp
plt.figure()
for CL, cluster_profile, bin_size in product(CLs, cl_profs, bin_sizes):
if (CL, cluster_profile) in exceptions: continue
if animal not in adat[CL.name] or session not in adat[
CL.name][animal]:
continue
cl_prof_name, _ = get_good_clusters(cluster_profile)
pts = []
try:
for K, correct_dp in adat[CL.name][animal][session][
cl_prof_name][bin_size][label].items():
if len(correct_dp) == 0: continue
y = 100.0 * np.sum(correct_dp > .5) / len(correct_dp)
pts.append((K, y))
except:
logging.warning('Something fishy with %s', CL.name)
if len(pts) == 0: continue
pts.sort(key=lambda x: x[0])
# Get the right color
CL_i = CLs.index(CL)
cp_i = cl_profs.index(cluster_profile)
b_i = bin_sizes.index(bin_size)
clr_i = CL_i * len(cl_profs) + cp_i
clr_str = clrs[clr_i] + ln_typs[b_i]
xs, ys = zip(*pts)
plt.plot(np.array(xs) * tpp,
ys,
clr_str,
label=CL.name,
linewidth=lw)
plt.legend(fontsize='x-small', loc='lower right')
plt.xlabel('Segment Length (s)')
plt.ylabel('Percent Correct')
#plt.title('Accuracy vs Segment Size, Animal %i Session %i'%(animal, session))
plt.ylim([60, 95])
plt.title('%i-Fold Validation' % (Folds, ))
plt.show()