dists_nc = squareform(0.5 * (1.0 - m.noise_corr))
m.linkage_nc = hac.linkage(dists_nc, method='complete')
m.hac_idxs_nc = hac.dendrogram(m.linkage_nc)['leaves']
hac_response_nc = m.data[:,m.hac_idxs_nc,:,:]
m.noise_corr_hac = noise_correlation(hac_response_nc)
# Cluster reliability.
# From signal correlations.
m.cl_idxs_sc = hac.fcluster(m.linkage_sc, n_clusters, criterion='maxclust')
m.sc_cluster_reliability = np.zeros((n_clusters+1,S))
m.mean_cluster_rsp_sc = np.zeros((S, n_clusters, L, R))
for i in range(1, n_clusters+1):
idxs = np.arange(N)
idxs = idxs[m.cl_idxs_sc == i]
m.mean_cluster_rsp_sc[:,i-1,:,:] = m.data[:,idxs,:,:].mean(axis=1)
m.sc_cluster_reliability = reliability(m.mean_cluster_rsp_sc)
# From noise correlations.
m.cl_idxs_nc = hac.fcluster(m.linkage_nc, n_clusters, criterion='maxclust')
m.nc_cluster_reliability = np.zeros((n_clusters+1,S))
m.mean_cluster_rsp_nc = np.zeros((S, n_clusters, L, R))
for i in range(1, n_clusters+1):
idxs = np.arange(N)
idxs = idxs[m.cl_idxs_nc == i]
m.mean_cluster_rsp_nc[:,i-1,:,:] = m.data[:,idxs,:,:].mean(axis=1)
m.nc_cluster_reliability = reliability(m.mean_cluster_rsp_nc)
############################################################################
# Plotting
if not os.path.isdir(os.path.join(PLOTS_DIR,'clustering')):
os.makedirs(os.path.join(PLOTS_DIR, 'clustering'))
for index in range(len(data)):
m = data[index]
print 'Mouse %s' % m.name
# Smoothing responses. Some unnecessarily clunky stuff here as well.
# Will do something about that later.
data[index] = smooth_responses(m)
m = data[index]
if exp_type == 'natural':
m.data = m.data[:5,:,:,:] # Remove K0, K1, etc.
S, N, L, R = m.data.shape
avg_response = np.mean(m.data, axis=(2,3))
m.reliability = reliability(m.data)
# Reliability-tuning curve correlation
if exp_type == 'grating':
m.tuning_curve = np.zeros((N,S))
m.rel_tc_corr = np.zeros(N)
for i in range(N):
m.tuning_curve[i,:] = avg_response[:,i]
r = m.reliability[i,:]
tc = m.tuning_curve[i,:]
r -= r.mean()
tc -= tc.mean()
m.rel_tc_corr[i] = np.dot(r,tc) / np.sqrt(np.dot(r,r)*np.dot(tc,tc))
############################################################################
# Plot