def get_predictor(
behavior: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray, float],
aligned_spike: Tuple[DataFrame, SparseRec], sample_rate: int,
spline: np.ndarray) -> Tuple[Predictors, np.ndarray, Grouping]:
"""Build Predictors from lever trajectory, trial parameters and spikes.
Args:
behavior: from behavior::get_behavior, amplitude, speed, delay, hit and reliability
reliability is one float, hit is a bool array with k trials. while amplitude, speed and delay are all floats
same length as number of hit trials.
aligned_spike: (spike, lever_trajectory) from decoder::align_xy, spike has n * k * t with n neurons, k trials
and t timepoints in a trial. lever trajectory has a value of k * t
Returns:
predictor_mat: a Predictor, with each of (event, period, trial, temporal) being 3D event and trial predictors,
n * k * t with n features, k trials and t timepoints
y: 3D neuronal activity, n * k * t with n neurons, k trials and t timepoints
grouping: a Predictor of 1d grouping arrays (event, period, trial, temporal)
"""
(amplitude, max_speed, delay, hit, _), (spike,
trace) = behavior, aligned_spike
trial_samples = int(round(sample_rate * 5.0))
delay_sample = np.minimum(trial_samples - 1,
np.rint(delay * sample_rate).astype(int))
delay_period = np.vstack(list(np.broadcast(0, delay_sample))).T
onset = np.rint(trace.timestamps * 5 / 256).astype(int)
trajectory: np.ndarray = take_segment(trace.values, onset, trial_samples)
speed: np.ndarray = take_segment(np.diff(trace.values), onset - 1,
trial_samples)
y = np.array([
take_segment(neuron, onset, trial_samples) for neuron in spike.values
])
preds = Predictors((0, delay_sample), [delay_period], [hit, delay],
[trajectory, speed])
grouping = Grouping([1, 2], [3], [4, 5], [6, 7])
preds, grouping = build_predictor(preds, grouping, hit, spline)
return preds, y, grouping
def neuron_lever(lever: SparseRec, neuron: DataFrame, neuron_rate: float,
motion_params: MotionParams) -> Tuple[DataFrame, np.ndarray]:
"""Read lever and corresponding neuron data by trial. I reimplement [fold_by] here to
acheive same length of the two segments.
Args:
data_file: file including spike and response(lever)
motion_params: MotionParams, a dict but all params are optional
may have [pre_time] and [post_time] in seconds, [quiet_var] and [event_thres] in float,
[window_size] in sample_no
Retursn:
neurons: a DataFrame with folded neurons
lever: a DataFrame with folded lever, resampled to neuron sample rate
"""
params = {
x: motion_params[x]
for x in ("quiet_var", "window_size", "event_thres")
if x in motion_params
}
event_onsets = event.find_response_onset(lever, **params)[0]
lever_resample = resample(lever.values[0], lever.sample_rate, neuron_rate)
anchor = np.rint(
(event_onsets / lever.sample_rate - motion_params['pre_time']) *
neuron_rate).astype(np.int)
duration = int(
round((motion_params.get('pre_time', 0.1) +
motion_params.get('post_time', 0.9)) * neuron_rate))
lever_folded = utils.take_segment(lever_resample, anchor, duration)
neuron_trials = np.stack([
utils.take_segment(trace, anchor, duration) for trace in neuron.values
])
mask, filtered = devibrate_trials(lever_folded, motion_params['pre_time'])
return neuron_trials[:, mask, :], lever_folded[mask, :]
def splice_trial_xy(xy: Tuple[DataFrame, SparseRec], sample_rate: float):
spike, trace = xy
onset = np.rint(trace.timestamps * 5 / 256).astype(int)
trial_samples = int(round(sample_rate * 5.0))
neurons = np.array([
take_segment(neuron, onset, trial_samples).ravel()
for neuron in spike.values
])
trajectory = take_segment(trace.values, onset, trial_samples).ravel()
timestamps = take_segment(trace.axes[0], onset, trial_samples).ravel()
return spike.create_like(neurons,
[spike.axes[0], timestamps]), trace.create_like(
trajectory, [timestamps])
def test_take_segment():
trace = np.arange(5, 1005)
points = np.array([3, 18, 518, 998])
result = take_segment(trace, points - 3, 5)
assert (np.array_equal(
result,
np.array([[5, 6, 7, 8, 9], [20, 21, 22, 23, 24],
[520, 521, 522, 523, 524], [1000, 1001, 1002, 1003, 1004]])))
def test_take_segments(lever_file, correct_last_trial):
raw = lever_file['response']['mvmtdata'].ravel()
onset = np.array(lever_file['response']['samples_start'] + [156000])
segments = take_segment(raw, onset, 1280)
print(segments)
assert (np.allclose(segments[-1, :], correct_last_trial))