)
_COVARIANCE_WITH_LARGE_NUMBERS_TEST = dict(
testcase_name='CovarianceWithLargeNumbers',
combiner=analyzers.CovarianceCombiner(output_shape=(2, 2),
numpy_dtype=np.float64),
batches=[
(np.array([[2e15, 0], [1e15, 0]]), ),
(np.array([[-2e15, 0], [-1e15, 0]]), ),
],
expected_outputs=[np.array([[2.5e30, 0], [0, 0]], dtype=np.float64)],
)
_PCA_WITH_DEGENERATE_COVARIANCE_MATRIX_TEST = dict(
testcase_name='PCAWithDegenerateCovarianceMatrix',
combiner=analyzers.PCACombiner(output_shape=(3, 3),
numpy_dtype=np.float64),
batches=[
(np.array([[0, 0, 1]]), ),
(np.array([[4, 0, 1], [2, -1, 1]]), ),
(np.array([[2, 1, 1]]), ),
],
expected_outputs=[
np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=np.float64)
],
)
def _make_mean_and_var_accumulator_from_instance(instance, axis=None):
return analyzers._WeightedMeanAndVarAccumulator(
count=np.sum(np.ones_like(instance), axis=axis),
mean=np.mean(instance, axis=axis),
],
)
_COVARIANCE_WITH_LARGE_NUMBERS_TEST = dict(
testcase_name='CovarianceWithLargeNumbers',
combiner=analyzers.CovarianceCombiner(numpy_dtype=np.float64),
batches=[
(np.array([[2e15, 0], [1e15, 0]]), ),
(np.array([[-2e15, 0], [-1e15, 0]]), ),
],
expected_outputs=[np.array([[2.5e30, 0], [0, 0]], dtype=np.float64)],
)
_PCA_WITH_DEGENERATE_COVARIANCE_MATRIX_TEST = dict(
testcase_name='PCAWithDegenerateCovarianceMatrix',
combiner=analyzers.PCACombiner(numpy_dtype=np.float64),
batches=[
(np.array([[0, 0, 1]]), ),
(np.array([[4, 0, 1], [2, -1, 1]]), ),
(np.array([[2, 1, 1]]), ),
],
expected_outputs=[
np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=np.float64)
],
)
def _make_mean_and_var_accumulator_from_instance(instance, axis=None):
return analyzers._MeanAndVarAccumulator(count=np.sum(
np.ones_like(instance), axis=axis),
mean=np.mean(instance, axis=axis),