def makeInputsAndOutputs(self, voxelsInEachDim, uniquebVals, numbVecs):
bVals, bVecs = generatebValsAndbVecs(uniquebVals, numbVecs)
gtab = gradient_table(bVals,
bVecs,
big_delta=self.bigDelta,
small_delta=self.smallDelta)
inputs = generateSyntheticInputs(
voxelsInEachDim,
gtab,
qMagnitudeTransform=self.qMagnitudeTransform)
outputs = generateSyntheticOutputsFromMultiTensorModel(
voxelsInEachDim, gtab, self.tensorEigenvalues, snr=None)
return inputs, outputs
def test_generateSyntheticInputs(self, mock_generateCoordinates):
qMagnitudes = np.array([0, 5])
bvecs = np.array([[0, 0, 0], [0, 0, 1]])
mock_gtab = mock.MagicMock(qvals=qMagnitudes, bvecs=bvecs)
(nx, ny, nz) = (1, 1, 2)
expectedCoordinates = np.array([[0, 0, 0], [0, 0, 1]])
expectedResult = np.vstack(
(np.r_[expectedCoordinates[0, :], qMagnitudes[0],
bvecs[0, :]], np.r_[expectedCoordinates[0, :],
qMagnitudes[1], bvecs[1, :]],
np.r_[expectedCoordinates[1, :], qMagnitudes[0],
bvecs[0, :]], np.r_[expectedCoordinates[1, :],
qMagnitudes[1], bvecs[1, :]]))
mock_generateCoordinates.return_value = expectedCoordinates
syntheticInputs = generateSyntheticInputs((nx, ny, nz), mock_gtab)
npt.assert_array_equal(syntheticInputs, expectedResult)
def generateSpatiallyCorrelatedData(self,
n=100,
lengthScale=1.,
scaling=1.,
noiseVariance=0.01):
voxelsInEachDim = (n, 1, 1)
bVals = np.array([0])
bVecs = np.array([[1, 0, 0]])
gtab = gradient_table(bVals,
bVecs,
big_delta=self.bigDelta,
small_delta=self.smallDelta)
inputs = generateSyntheticInputs(voxelsInEachDim, gtab)
xCoordinate = inputs[:, 0]
distances = np.abs(xCoordinate[:, np.newaxis] -
xCoordinate[np.newaxis, :])
spatialCovariance = (np.exp(-scaling * (distances / lengthScale)**2) +
noiseVariance * np.eye(n))
outputs = scaling * np.random.multivariate_normal(
np.zeros(n, ), spatialCovariance)
outputs = outputs[:, np.newaxis]
return inputs, outputs
def test_dataGeneration(self):
voxelsInEachDim = (2, 3, 4)
bvals = np.concatenate([[0], 1500 * np.ones(6), 3000 * np.ones(6)])
sq2 = np.sqrt(2) / 2
setOfbvecs = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1], [sq2, sq2, 0],
[sq2, 0, sq2], [0, sq2, sq2]])
bvecs = np.vstack([np.array([[0, 0, 0]]), setOfbvecs, setOfbvecs])
gtab = gradient_table(bvals,
bvecs,
big_delta=self.bigDelta,
small_delta=self.smallDelta)
inputs = generateSyntheticInputs(voxelsInEachDim, gtab)
outputsLatent = generateSyntheticOutputsFromMultiTensorModel(
voxelsInEachDim, gtab, self.tensorEigenvalues, snr=None)
self.assertTrue(
np.min(outputsLatent) >= 0 and np.max(outputsLatent) <= 1)
totalNumberOfSamples = np.prod(voxelsInEachDim) * len(bvals)
npt.assert_array_equal(inputs.shape, (totalNumberOfSamples, 7),
'Input shapes don\'t match')
npt.assert_array_equal(outputsLatent.shape, (totalNumberOfSamples, 1),
'Output shapes don\'t match')