def test_dichotomized_gaussian(self):
bin_means = np.array([7, 9])
bin_cov = np.array([[7, 3], [3, 9]])
num_samples = 10000
np.random.seed(42)
# calculate marginal distribution of Poisson
pmfs, cmfs, supports = poisson_marginals(bin_means)
self.assertEqual(len(pmfs), 2)
self.assertTrue(sum(map(sum, pmfs)) - 2. < 1e-4)
self.assertEqual(len(pmfs[0]), len(supports[0]))
self.assertEqual(len(pmfs[1]), len(supports[1]))
self.assertEqual(len(pmfs[0]), len(cmfs[0]))
self.assertEqual(len(pmfs[1]), len(cmfs[1]))
# find paramters of DG
gauss_means, gauss_covs, joints = find_dg_any_marginal(pmfs, bin_cov, supports)
# generate samples
np.random.seed(0)
samples, hists = sample_dg_any_marginal(gauss_means, gauss_covs, num_samples, supports)
self.assertTrue(samples[:, 0].mean() - 7, 1e-2)
self.assertTrue(samples[:, 1].mean() - 9, 1e-2)
def sample_from_model(self, trials=None, reshape=False):
"""
Missing documentation
Parameters
----------
trials : Type, optional
Description (default None)
reshape : bool, optional
Description (default False)
Returns
-------
Value : Type
Description
"""
trials = trials or range(self._original_spikes.T)
spikes_windowed = self._original_spikes.to_windowed(
self._window_size, trials)
X = np.zeros(
(len(trials), self._original_spikes.N, self._original_spikes.M))
# statistics
for c, t in enumerate(trials):
bin_means = spikes_windowed.spikes[t, :, :].mean(axis=1)
bin_cov = np.cov(spikes_windowed.spikes[t, :, :])
# calculate marginal distribution of Poisson
pmfs, cmfs, supports = poisson_marginals(bin_means)
# find paramters of DG
gauss_means, gauss_cov, joints = find_dg_any_marginal(
pmfs, bin_cov, supports)
# generate samples
samples, hists = sample_dg_any_marginal(gauss_means, gauss_cov,
self._original_spikes.M,
supports)
X[c, :, :] = samples.T
if reshape:
Y = np.zeros((X.shape[0] * X.shape[2], X.shape[1]))
tot = 0
for t in xrange(len(trials)):
for c in xrange(X.shape[2]):
Y[tot, :] = X[t, :, c]
tot += 1
return Y
return Spikes(spikes=X)
def sample_from_model(self, trials=None, reshape=False):
"""
Missing documentation
Parameters
----------
trials : Type, optional
Description (default None)
reshape : bool, optional
Description (default False)
Returns
-------
Value : Type
Description
"""
trials = trials or range(self._original_spikes.T)
spikes_windowed = self._original_spikes.to_windowed(self._window_size, trials)
X = np.zeros((len(trials), self._original_spikes.N, self._original_spikes.M))
# statistics
for c, t in enumerate(trials):
bin_means = spikes_windowed.spikes[t, :, :].mean(axis=1)
bin_cov = np.cov(spikes_windowed.spikes[t, :, :])
# calculate marginal distribution of Poisson
pmfs, cmfs, supports = poisson_marginals(bin_means)
# find paramters of DG
gauss_means, gauss_cov, joints = find_dg_any_marginal(pmfs, bin_cov, supports)
# generate samples
samples, hists = sample_dg_any_marginal(gauss_means, gauss_cov, self._original_spikes.M, supports)
X[c, :, :] = samples.T
if reshape:
Y = np.zeros((X.shape[0] * X.shape[2], X.shape[1]))
tot = 0
for t in xrange(len(trials)):
for c in xrange(X.shape[2]):
Y[tot, :] = X[t, :, c]
tot += 1
return Y
return Spikes(spikes=X)