def test_compute_covariance_matrix():
kk = generate_simple_test_kk(points_for_cluster_mask=0.5)
num_cluster_members = kk.num_cluster_members
num_clusters = kk.num_clusters_alive
num_features = kk.num_features
cluster_mask_sum = zeros((num_clusters, num_features))
for cluster in range(kk.num_special_clusters, num_clusters):
accumulate_cluster_mask_sum(kk, cluster_mask_sum[cluster, :],
kk.get_spikes_in_cluster(cluster))
cluster_mask_sum[:kk.
num_special_clusters, :] = -1 # ensure that special clusters are masked
cov_matrices = []
for cluster in range(1, num_clusters):
cluster_mean = compute_cluster_mean(kk, cluster)
curmask = cluster_mask_sum[cluster, :]
unmasked, = (curmask >= kk.points_for_cluster_mask).nonzero()
masked, = (curmask < kk.points_for_cluster_mask).nonzero()
unmasked = array(unmasked, dtype=int)
masked = array(masked, dtype=int)
cov = BlockPlusDiagonalMatrix(masked, unmasked)
compute_covariance_matrix(kk, cluster, cluster_mean, cov)
f2m = kk.orig_features - cluster_mean[newaxis, :]
ct = kk.orig_correction_terms
spikes = kk.get_spikes_in_cluster(cluster)
block = zeros_like(cov.block)
unmasked = cov.unmasked
for spike in spikes:
if cluster == 1:
for i in range(len(unmasked)):
block[i, i] += f2m[spike, unmasked[i]]**2
else:
block[:, :] += f2m[spike, unmasked][:, newaxis] * f2m[
spike, unmasked][newaxis, :]
for i in range(len(unmasked)):
block[i, i] += ct[spike, unmasked[i]]
if cluster == 1:
point = kk.mua_point
else:
point = kk.noise_point
for i in range(len(cov.unmasked)):
block[i, i] += point * kk.data.noise_variance[cov.unmasked[i]]
factor = 1.0 / (num_cluster_members[cluster] + point - 1)
block *= factor
assert_array_almost_equal(block, cov.block)
cov_matrices.append(deepcopy(cov))
return cov_matrices
def test_compute_covariance_matrix():
kk = generate_simple_test_kk(points_for_cluster_mask=0.5)
num_cluster_members = kk.num_cluster_members
num_clusters = kk.num_clusters_alive
num_features = kk.num_features
cluster_mask_sum = zeros((num_clusters, num_features))
for cluster in range(kk.num_special_clusters, num_clusters):
accumulate_cluster_mask_sum(kk, cluster_mask_sum[cluster, :], kk.get_spikes_in_cluster(cluster))
cluster_mask_sum[:kk.num_special_clusters, :] = -1 # ensure that special clusters are masked
cov_matrices = []
for cluster in range(1, num_clusters):
cluster_mean = compute_cluster_mean(kk, cluster)
curmask = cluster_mask_sum[cluster, :]
unmasked, = (curmask>=kk.points_for_cluster_mask).nonzero()
masked, = (curmask<kk.points_for_cluster_mask).nonzero()
unmasked = array(unmasked, dtype=int)
masked = array(masked, dtype=int)
cov = BlockPlusDiagonalMatrix(masked, unmasked)
compute_covariance_matrix(kk, cluster, cluster_mean, cov)
f2m = kk.orig_features-cluster_mean[newaxis, :]
ct = kk.orig_correction_terms
spikes = kk.get_spikes_in_cluster(cluster)
block = zeros_like(cov.block)
unmasked = cov.unmasked
for spike in spikes:
if cluster==1:
for i in range(len(unmasked)):
block[i, i] += f2m[spike, unmasked[i]]**2
else:
block[:, :] += f2m[spike, unmasked][:, newaxis]*f2m[spike, unmasked][newaxis, :]
for i in range(len(unmasked)):
block[i, i] += ct[spike, unmasked[i]]
if cluster==1:
point = kk.mua_point
else:
point = kk.noise_point
for i in range(len(cov.unmasked)):
block[i, i] += point*kk.data.noise_variance[cov.unmasked[i]]
factor = 1.0/(num_cluster_members[cluster]+point-1)
block *= factor
assert_array_almost_equal(block, cov.block)
cov_matrices.append(deepcopy(cov))
return cov_matrices
def test_accumulate_cluster_mask_sum():
kk = generate_simple_test_kk()
fet = kk.orig_fet
fmask = kk.orig_fmask
assert kk.num_clusters_alive==5
num_clusters = kk.num_clusters_alive
num_features = kk.num_features
cluster_mask_sum = zeros((num_clusters, num_features))
cluster_mask_sum[:2, :] = -1 # ensure that clusters 0 and 1 are masked
for cluster in range(2, num_clusters):
accumulate_cluster_mask_sum(kk, cluster_mask_sum[cluster, :], kk.get_spikes_in_cluster(cluster))
# cluster 2 has only point 0 in it, so the cluster_mask sum should be just the corresponding
# fmask line
assert_array_almost_equal(cluster_mask_sum[2, :], fmask[0, :])
# similarly for the others
assert_array_almost_equal(cluster_mask_sum[3, :], fmask[1, :]+fmask[2, :])
assert_array_almost_equal(cluster_mask_sum[4, :], fmask[3, :])
assert (cluster_mask_sum[0, :]==-1).all()
assert (cluster_mask_sum[1, :]==-1).all()