def test_preprocess_spike_train(self):
stimulus = arange(20).T.reshape(-1, 2).T
spike_train = arange(10).reshape(1, -1)
spike_history = 3
stimulus_history = 2
stimuli = extract_windows(stimulus, stimulus_history)
spikes = extract_windows(spike_train, spike_history)
stimuli = stimuli[:, -spikes.shape[1]:]
spikes = spikes[:, -stimuli.shape[1]:]
self.assertEqual(stimuli.shape[0],
stimulus.shape[0] * stimulus_history)
self.assertEqual(
stimuli.shape[1],
stimulus.shape[1] - max([stimulus_history, spike_history]) + 1)
self.assertEqual(spikes.shape[0], spike_train.shape[0] * spike_history)
self.assertEqual(
spikes.shape[1],
spike_train.shape[1] - max([stimulus_history, spike_history]) + 1)
self.assertLess(
max(abs(spikes[:, -1] - spike_train[0, -spike_history:])), 1e-8)
self.assertLess(
max(abs(stimuli[:, -1] -
stimulus[:, -stimulus_history:].T.ravel())), 1e-8)
def generate_data_from_spike_train(stimulus, stimulus_history, spike_train=None, spike_history=0):
"""
Extracts windows from a stimulus time-series and a spike train.
@type stimulus: C{ndarray}
@param stimulus: time-series (NxT) where the second dimension is time
@type stimulus_history: C{int}
@param stimulus_history: length of extracted stimulus windows
@type spike_train: C{ndarray}
@param spike_train: spikes corresponding to stimulus (1xT)
@type spike_history: C{int}
@param spike_history: length of extracted spike windows
@rtype: C{tuple}/C{ndarray}
@return: stimulus windows, spike histories, and spikes
"""
if stimulus.ndim == 1:
stimulus = stimulus.reshape(1, -1)
if spike_train is None:
return extract_windows(stimulus, stimulus_history)
if spike_train.ndim == 1:
spike_train = spike_train.reshape(1, -1)
if stimulus.shape[1] != spike_train.shape[1]:
raise ValueError('Stimulus and spike train should have the same length.')
# extract stimulus and spike history windows
spikes = extract_windows(spike_train, spike_history + 1)
stimuli = extract_windows(stimulus, stimulus_history)
# make sure stimuli and spikes are aligned
spikes = spikes[:, -stimuli.shape[1]:]
stimuli = stimuli[:, -spikes.shape[1]:]
# separate last spike of each window
outputs = spikes[[-1]]
spikes = spikes[:-1]
if spike_history > 0:
return stimuli, spikes, outputs
else:
return stimuli, outputs
def test_preprocess_spike_train(self): stimulus = arange(20).T.reshape(-1, 2).T spike_train = arange(10).reshape(1, -1) spike_history = 3 stimulus_history = 2 stimuli = extract_windows(stimulus, stimulus_history) spikes = extract_windows(spike_train, spike_history) stimuli = stimuli[:, -spikes.shape[1]:] spikes = spikes[:, -stimuli.shape[1]:] self.assertEqual(stimuli.shape[0], stimulus.shape[0] * stimulus_history) self.assertEqual(stimuli.shape[1], stimulus.shape[1] - max([stimulus_history, spike_history]) + 1) self.assertEqual(spikes.shape[0], spike_train.shape[0] * spike_history) self.assertEqual(spikes.shape[1], spike_train.shape[1] - max([stimulus_history, spike_history]) + 1) self.assertLess(max(abs(spikes[:, -1] - spike_train[0, -spike_history:])), 1e-8) self.assertLess(max(abs(stimuli[:, -1] - stimulus[:, -stimulus_history:].T.ravel())), 1e-8)
