def test_dct_backend(self, xp, dtype):
backend = 'scipy' if xp is np else cp_fft
with scipy_fft.set_backend(backend):
fft_func = getattr(scipy_fft, self.function)
r = xp.logspace(-2, 2, 10)
dln = xp.log(r[1]/r[0])
return fft_func(r, dln, mu=0.5)
def preprocess(data,scale_min=-1,scale_max=1,training_min=None, training_max=None,
sigma=(0,2,2), eps=0.01,ST_method='DCT',cpus_to_use =-1,inverse=False,**kwargs):
"""
Pre-processing for data sets.
Each data set (train, val, pred)
Must be processed separately.
To avoid bias, only training
set must be used to estimate
training parameters
"""
if ST_method == 'DCT':
ST = ST_ndim_DCT
elif ST_method == 'FFT':
ST = ST_ndim_FFT
with set_backend(pyfftw.interfaces.scipy_fft), pyfftw.interfaces.scipy_fft.set_workers(cpus_to_use):
#faster if we enable cache using pyfftw
pyfftw.interfaces.cache.enable()
# perform standardizing transform using frequency method of your choice
data = ST(data,sigma,eps,inverse)
if (training_min is None) or (training_max is None):
# Assumes training set
training_min, training_max = data.min(), data.max()
data = scale(data,scale_min,scale_max,training_min,training_max,inverse)
return data, training_min, training_max
else:
# Appropriate for val/test set
data = scale(data,scale_min,scale_max,training_min,training_max,inverse)
return data
def test_backend_call(func, np_func, mock):
x = np.arange(20).reshape((10, 2))
answer = np_func(x)
assert_allclose(func(x), answer, atol=1e-10)
with set_backend(mock_backend, only=True):
mock.number_calls = 0
y = func(x)
assert_equal(y, mock.return_value)
assert_equal(mock.number_calls, 1)
assert_allclose(func(x), answer, atol=1e-10)
def test_backend_plan(func, mock):
x = np.arange(20).reshape((10, 2))
with pytest.raises(NotImplementedError, match='precomputed plan'):
func(x, plan='foo')
with set_backend(mock_backend, only=True):
mock.number_calls = 0
y = func(x, plan='foo')
assert_equal(y, mock.return_value)
assert_equal(mock.number_calls, 1)
assert_equal(mock.last_args[1]['plan'], 'foo')
def test_dct_backend(self, xp, dtype):
backend = 'scipy' if xp is np else cp_fft
with scipy_fft.set_backend(backend):
fft_func = getattr(scipy_fft, self.function)
return self._run_transform(fft_func, xp, dtype)
