def test_revcorr_raises():
"""Test raising ValueErrors with incorrect inputs"""
np.random.seed(0)
with pytest.raises(ValueError):
flt.revcorr(np.random.randn(10, ), np.random.randn(10, 2), 2)
with pytest.raises(ValueError):
flt.revcorr(np.random.randn(10, 2), np.random.randn(10, 3), 2)
def test_revcorr_acausal():
"""Test computation of a 1D linear filter by reverse correlation,
including acausal lag values.
"""
np.random.seed(0)
# Create fake filter, 100 time points
nbefore, nafter = 90, 10
filter_length = nbefore + nafter
true_filter = utils.create_temporal_filter(filter_length)
# Compute linear response
stim_length = 10000
stimulus = np.random.randn(stim_length,)
response = flt.linear_response(true_filter, stimulus)
# Reverse correlation, pad response with zeros to so that
# stim and response match. These will be ignored by
# filtertools.revcorr anyway.
padded_response = np.concatenate((np.zeros((filter_length - 1,)),
response), axis=0)
filt = flt.revcorr(stimulus, padded_response, nbefore, nafter)[0]
filt /= np.linalg.norm(filt)
tol = 0.1
assert np.allclose(true_filter, filt, atol=tol)
def test_revcorr_1d_ignores_beginning():
"""Verify revcorr ignores the first filter-length points of the stimulus,
to only consider those points which the response and stimulus overlap
completely.
"""
filt = np.array(((1, 0, 0)))
stim = np.concatenate(((1, ), np.zeros((10, ))), axis=0)
response = np.convolve(filt, stim, 'full')[:stim.size]
recovered, lags = flt.revcorr(stim, response, filt.size)
assert np.allclose(recovered, 0)
def test_revcorr_1d_ignores_beginning():
"""Verify revcorr ignores the first filter-length points of the stimulus,
to only consider those points which the response and stimulus overlap
completely.
"""
filt = np.array(((1, 0, 0)))
stim = np.concatenate(((1,), np.zeros((10,))), axis=0)
response = np.convolve(filt, stim, 'full')[:stim.size]
recovered, lags = flt.revcorr(stim, response, filt.size)
assert np.allclose(recovered, 0)
def test_revcorr_nd():
"""Test computation of 3D linear filter by reverse correlation.
The reverse correlation should return the time-reverse of the
linear filter, scaled by the number of spatial points.
"""
ndim = 3
filt = np.zeros((3, ) + ((2, ) * ndim))
filt[0] = 1.
stim = np.zeros((10, ) + ((2, ) * ndim))
stim[5] = 1.
response = flt.linear_response(filt, stim)
recovered, lags = flt.revcorr(stim, response, filt.shape[0])
assert np.allclose(recovered[-1], filt[0].sum())
assert np.allclose(recovered[:-1], 0)
def test_revcorr_1d():
"""Test computation of 1D reverse correlation.
The reverse-correlation should recover the time-reverse of the
linear filter, and the lags should be start at negative values
and be strictly increasing.
"""
filt = np.array(((1, 0, 0)))
stim = np.zeros((10, ))
stim[5] = 1
response = np.convolve(filt, stim, 'full')[:stim.size]
recovered, lags = flt.revcorr(stim, response, filt.size)
assert np.allclose(recovered, filt[::-1])
assert lags[0] == -(filt.size - 1)
assert (np.diff(lags) == 1).all()
def test_revcorr_nd():
"""Test computation of 3D linear filter by reverse correlation.
The reverse correlation should return the time-reverse of the
linear filter, scaled by the number of spatial points.
"""
ndim = 3
filt = np.zeros((3,) + ((2,) * ndim))
filt[0] = 1.
stim = np.zeros((10,) + ((2,) * ndim))
stim[5] = 1.
response = flt.linear_response(filt, stim)
recovered, lags = flt.revcorr(stim, response, filt.shape[0])
assert np.allclose(recovered[-1], filt[0].sum())
assert np.allclose(recovered[:-1], 0)
def test_revcorr_1d():
"""Test computation of 1D reverse correlation.
The reverse-correlation should recover the time-reverse of the
linear filter, and the lags should be start at negative values
and be strictly increasing.
"""
filt = np.array(((1, 0, 0)))
stim = np.zeros((10,))
stim[5] = 1
response = np.convolve(filt, stim, 'full')[:stim.size]
recovered, lags = flt.revcorr(stim, response, filt.size)
assert np.allclose(recovered, filt[::-1])
assert lags[0] == -(filt.size - 1)
assert (np.diff(lags) == 1).all()
def test_revcorr_acausal():
"""Test computation of a 1D linear filter by reverse correlation,
including acausal lag values. The reverse-correlation should recover
the time-reverse of the linear filter.
"""
filt = np.array(((1, 0, 0)))
stim = np.zeros((10, ))
stim[5] = 1.0
response = np.convolve(filt, stim, 'full')[:stim.size]
nsamples_after = 2
recovered, lags = flt.revcorr(stim, response, filt.size, nsamples_after)
assert np.allclose(recovered[nsamples_after:], filt[::-1])
assert np.allclose(recovered[:filt.size], 0)
assert lags[0] == -(filt.size - 1)
assert lags[-1] == nsamples_after
assert (np.diff(lags) == 1).all()
def test_revcorr_acausal():
"""Test computation of a 1D linear filter by reverse correlation,
including acausal lag values. The reverse-correlation should recover
the time-reverse of the linear filter.
"""
filt = np.array(((1, 0, 0)))
stim = np.zeros((10,))
stim[5] = 1.0
response = np.convolve(filt, stim, 'full')[:stim.size]
nsamples_after = 2
recovered, lags = flt.revcorr(stim, response, filt.size, nsamples_after)
assert np.allclose(recovered[nsamples_after:], filt[::-1])
assert np.allclose(recovered[:filt.size], 0)
assert lags[0] == -(filt.size - 1)
assert lags[-1] == nsamples_after
assert (np.diff(lags) == 1).all()
def test_revcorr_nd():
np.random.seed(0)
"""Test computation of 3D linear filter by reverse correlation"""
# Create fake filter
filter_length = 100
nx, ny = 10, 10
true_filter = utils.create_spatiotemporal_filter(nx, ny, filter_length)[-1]
# Compute linear response
stim_length = 10000
stimulus = np.random.randn(stim_length, nx, ny)
response = flt.linear_prediction(true_filter, stimulus)
# Reverse correlation
filt = flt.revcorr(response, stimulus, filter_length)
filt /= np.linalg.norm(filt)
tol = 0.1
assert np.allclose(true_filter, filt, atol=tol)
def test_revcorr_1d():
"""Test computation of a 1D linear filter by reverse correlation"""
np.random.seed(0)
# Create fake filter, 100 time points
filter_length = 100
true_filter = utils.create_temporal_filter(filter_length)
# Compute linear response
stim_length = 10000
stimulus = np.random.randn(stim_length, )
response = flt.linear_prediction(true_filter, stimulus)
# Reverse correlation
filt = flt.revcorr(response, stimulus, filter_length)
filt /= np.linalg.norm(filt)
tol = 0.1
assert np.allclose(true_filter, filt, atol=tol)
def test_revcorr_nd():
np.random.seed(0)
"""Test computation of 3D linear filter by reverse correlation"""
# Create fake filter
filter_length = 100
nx, ny = 10, 10
true_filter = utils.create_spatiotemporal_filter(nx, ny, filter_length)[-1]
# Compute linear response
stim_length = 10000
stimulus = np.random.randn(stim_length, nx, ny)
response = flt.linear_response(true_filter, stimulus)
# Reverse correlation, pad response with zeros to so that
# stim and response match. These will be ignored by
# filtertools.revcorr anyway.
padded_response = np.concatenate((np.zeros((filter_length - 1,)),
response), axis=0)
filt = flt.revcorr(stimulus, padded_response, filter_length)[0]
filt /= np.linalg.norm(filt)
tol = 0.1
assert np.allclose(true_filter, filt, atol=tol)
def test_revcorr_raises():
"""Test raising ValueErrors with incorrect inputs"""
with pytest.raises(ValueError):
flt.revcorr(np.zeros((10, 1)), np.zeros((11, )), 2)[0]
with pytest.raises(ValueError):
flt.revcorr(np.zeros((10, 3)), np.zeros((10, 2)), 2)[0]
def test_revcorr_raises():
"""Test raising ValueErrors with incorrect inputs"""
with pytest.raises(ValueError):
flt.revcorr(np.zeros((10, 1)), np.zeros((11,)), 2)[0]
with pytest.raises(ValueError):
flt.revcorr(np.zeros((10, 3)), np.zeros((10, 2)), 2)[0]
