def test_strf():
n_spatial_basis = 36
n_temporal_basis = 7
patch_size = 50
# Instantiate strf object
strf_ = STRF(patch_size=patch_size, sigma=5,
n_spatial_basis=n_spatial_basis,
n_temporal_basis=n_temporal_basis)
# Design a spatial basis
spatial_basis = strf_.make_cosine_basis()
assert_equal(len(spatial_basis), 2)
for basis in spatial_basis:
assert_equal(basis.shape[0], patch_size)
assert_equal(basis.shape[1], patch_size)
spatial_basis = strf_.make_gaussian_basis()
assert_equal(len(spatial_basis), n_spatial_basis)
# Visualize spatial basis
strf_.visualize_gaussian_basis(spatial_basis)
# Design temporal basis
time_points = np.linspace(-100., 100., 10.)
centers = [-75., -50., -25., 0, 25., 50., 75.]
width = 10.
temporal_basis = strf_.make_raised_cosine_temporal_basis(
time_points=time_points,
centers=centers,
widths=width * np.ones(7))
assert_equal(temporal_basis.shape[0], len(time_points))
assert_equal(temporal_basis.shape[1], n_temporal_basis)
# Project to spatial basis
I = np.zeros(shape=(patch_size, patch_size))
row = 5
col = 10
I[row, col] = 1
basis_projection = strf_.project_to_spatial_basis(I, spatial_basis)
assert_equal(len(basis_projection), n_spatial_basis)
# Recover image from basis projection
weights = np.random.normal(size=n_spatial_basis)
RF = strf_.make_image_from_spatial_basis(spatial_basis, weights)
assert_equal(RF.shape[0], patch_size)
assert_equal(RF.shape[1], patch_size)
# Convolve with temporal basis
n_samples = 100
n_features = n_spatial_basis
design_matrix = np.random.normal(size=(n_samples, n_features))
features = strf_.convolve_with_temporal_basis(
design_matrix, temporal_basis)
assert_equal(features.shape[0], n_samples)
assert_equal(features.shape[1], n_features * n_temporal_basis)
# Design prior covariance
PriorCov = strf_.design_prior_covariance(
sigma_temporal=3., sigma_spatial=5.)
assert_equal(PriorCov.shape[0], PriorCov.shape[1])
assert_equal(PriorCov.shape[0], n_spatial_basis * n_temporal_basis)
########################################################
from pyglmnet import utils
n_samples = Xtrain.shape[0]
Tau = utils.tikhonov_from_prior(prior_cov, n_samples)
glm = GLMCV(distr='poisson', alpha=0., Tau=Tau, score_metric='pseudo_R2', cv=3)
glm.fit(Xtrain, Ytrain)
print("train score: %f" % glm.score(Xtrain, Ytrain))
print("test score: %f" % glm.score(Xtest, Ytest))
weights = glm.beta_
########################################################
# Visualize
for time_bin_ in range(n_temporal_basis):
RF = strf_model.make_image_from_spatial_basis(
spatial_basis,
weights[range(time_bin_, n_spatial_basis * n_temporal_basis,
n_temporal_basis)])
plt.subplot(1, n_temporal_basis, time_bin_ + 1)
plt.imshow(RF, cmap='Blues', interpolation='none')
titletext = str(centers[time_bin_])
plt.title(titletext)
plt.axis('off')
plt.show()
########################################################
def test_strf():
n_spatial_basis = 36
n_temporal_basis = 7
patch_size = 50
# Instantiate strf object
strf_ = STRF(patch_size=patch_size,
sigma=5,
n_spatial_basis=n_spatial_basis,
n_temporal_basis=n_temporal_basis)
# Design a spatial basis
spatial_basis = strf_.make_cosine_basis()
assert_equal(len(spatial_basis), 2)
for basis in spatial_basis:
assert_equal(basis.shape[0], patch_size)
assert_equal(basis.shape[1], patch_size)
spatial_basis = strf_.make_gaussian_basis()
assert_equal(len(spatial_basis), n_spatial_basis)
# Visualize spatial basis
strf_.visualize_gaussian_basis(spatial_basis)
# Design temporal basis
time_points = np.linspace(-100., 100., 10.)
centers = [-75., -50., -25., 0, 25., 50., 75.]
width = 10.
temporal_basis = strf_.make_raised_cosine_temporal_basis(
time_points=time_points, centers=centers, widths=width * np.ones(7))
assert_equal(temporal_basis.shape[0], len(time_points))
assert_equal(temporal_basis.shape[1], n_temporal_basis)
# Project to spatial basis
I = np.zeros(shape=(patch_size, patch_size))
row = 5
col = 10
I[row, col] = 1
basis_projection = strf_.project_to_spatial_basis(I, spatial_basis)
assert_equal(len(basis_projection), n_spatial_basis)
# Recover image from basis projection
weights = np.random.normal(size=n_spatial_basis)
RF = strf_.make_image_from_spatial_basis(spatial_basis, weights)
assert_equal(RF.shape[0], patch_size)
assert_equal(RF.shape[1], patch_size)
# Convolve with temporal basis
n_samples = 100
n_features = n_spatial_basis
design_matrix = np.random.normal(size=(n_samples, n_features))
features = strf_.convolve_with_temporal_basis(design_matrix,
temporal_basis)
assert_equal(features.shape[0], n_samples)
assert_equal(features.shape[1], n_features * n_temporal_basis)
# Design prior covariance
PriorCov = strf_.design_prior_covariance(sigma_temporal=3.,
sigma_spatial=5.)
assert_equal(PriorCov.shape[0], PriorCov.shape[1])
assert_equal(PriorCov.shape[0], n_spatial_basis * n_temporal_basis)