Merge branch 'misc_tweaks' into 'master'

# Conflicts:
#   nifty/sugar.py

.gitlab-ci.yml

@@ -16,38 +16,21 @@ before_script:
   - pip install --upgrade -r ci/requirements.txt
 
 
-test_min:
+test_all:
   stage: test
   script:
     - python setup.py build_ext --inplace
-    - nosetests -vv
-
-test_mpi:
-  stage: test
-  script:
+    - nosetests
     - ci/install_pyHealpix.sh
     - ci/install_mpi4py.sh
     - python setup.py build_ext --inplace
-    - nosetests -vv
-
-test_mpi_fftw:
-  stage: test
-  script:
-    - ci/install_pyHealpix.sh
-    - ci/install_mpi4py.sh
+    - nosetests
     - ci/install_pyfftw.sh
     - python setup.py build_ext --inplace
-    - nosetests -vv
-
-test_mpi_fftw_hdf5:
-  stage: test
-  script:
-    - ci/install_pyHealpix.sh
-    - ci/install_mpi4py.sh
-    - ci/install_pyfftw.sh
+    - nosetests
     - ci/install_h5py.sh
     - python setup.py build_ext --inplace
-    - nosetests -vv --with-coverage --cover-package=nifty --cover-branches
+    - nosetests --with-coverage --cover-package=nifty --cover-branches
     - >
       coverage report | grep TOTAL | awk '{ print "TOTAL: "$6; }'
 

README.md

@@ -11,7 +11,7 @@ Summary
 
 ### Description
 
-**NIFTY**, "**N**umerical **I**nformation **F**ield **T**heor**y**", is
+**NIFTY**, "**N**umerical **I**nformation **F**ield **T**heor<strong>y</strong>", is
 a versatile library designed to enable the development of signal
 inference algorithms that operate regardless of the underlying spatial
 grid and its resolution. Its object-oriented framework is written in
@@ -87,21 +87,17 @@ Starting with a fresh Ubuntu installation move to a folder like
 
 -   Install basic packages like python, python-dev, gsl and others:
 
-        sudo apt-get install curl git autoconf python-dev python-pip python-numpy
+        sudo apt-get install curl git autoconf libtool python-dev python-pip python-numpy
 
 -   Install pyHealpix:
 
         git clone https://gitlab.mpcdf.mpg.de/ift/pyHealpix.git
-        cd pyHealpix
-        autoreconf -i && ./configure --prefix=$HOME/.local --enable-openmp --enable-native-optimizations && make -j4 && make install
-        cd ..
+        (cd pyHealpix && autoreconf -i && ./configure --prefix=$HOME/.local --enable-openmp --enable-native-optimizations && make -j4 install)
 
 -   Finally, NIFTy:
 
         git clone https://gitlab.mpcdf.mpg.de/ift/NIFTy.git
-        cd NIFTy
-        python setup.py install --user
-        cd ..
+        (cd NIFTy && python setup.py install --user)
 
 ### Installation on Linux systems in general
 
@@ -139,16 +135,13 @@ may cause trouble.
 -   Install NIFTy:
 
         git clone https://gitlab.mpcdf.mpg.de/ift/NIFTy.git
-        cd NIFTy
-        python setup.py install --user
-        cd ..
+        (cd NIFTy && python setup.py install --user)
 
 ### Running the tests
 
 In oder to run the tests one needs two additional packages:
 
-    pip install nose
-    pip install parameterized
+    pip install nose parameterized
 
 Afterwards the tests (including a coverage report) are run using the following
 command in the repository root:

ci/install_basics.sh

 #!/bin/bash
 
-apt-get install -y build-essential python python-pip python-dev git autoconf gsl-bin libgsl-dev wget 
+apt-get install -y build-essential python python-pip python-dev git autoconf libtool gsl-bin libgsl-dev wget

ci/install_h5py.sh

 #!/bin/bash
 
 apt-get install -y libhdf5-10 libhdf5-dev libhdf5-openmpi-10 libhdf5-openmpi-dev hdf5-tools
-
-export CC=mpicc
-export HDF5_DIR=/usr/lib/x86_64-linux-gnu/hdf5/openmpi
-export HDF5_MPI="ON"
-
-pip install --no-binary=h5py h5py
+CC=mpicc HDF5_DIR=/usr/lib/x86_64-linux-gnu/hdf5/openmpi HDF5_MPI="ON" pip install --no-binary=h5py h5py

ci/install_mpi4py.sh

 #!/bin/bash
 
 apt-get install -y openmpi-bin libopenmpi-dev
-pip install mpi4py
\ No newline at end of file
+pip install mpi4py

ci/install_pyHealpix.sh

 #!/bin/bash
 
 git clone https://gitlab.mpcdf.mpg.de/ift/pyHealpix.git
-cd pyHealpix
-autoreconf -i && ./configure && make -j4 install
-cd ..
+(cd pyHealpix && autoreconf -i && ./configure --enable-openmp --enable-native-optimizations && make -j4 install)
 rm -rf pyHealpix

ci/install_pyfftw.sh

@@ -3,7 +3,5 @@
 apt-get install -y libatlas-base-dev libfftw3-bin libfftw3-dev libfftw3-double3 libfftw3-long3 libfftw3-mpi-dev libfftw3-mpi3 libfftw3-quad3 libfftw3-single3
 
 git clone -b mpi https://github.com/fredros/pyFFTW.git
-cd pyFFTW/
-CC=mpicc python setup.py build_ext install
-cd ..
-rm -r pyFFTW
+(cd pyFFTW && CC=mpicc python setup.py build_ext install)
+rm -rf pyFFTW

nifty/field_types/field_array.py

@@ -22,12 +22,11 @@ from field_type import FieldType
 class FieldArray(FieldType):
 
     def __init__(self, shape):
+        super(FieldArray, self).__init__()
         try:
-            new_shape = tuple([int(i) for i in shape])
+            self._shape = tuple([int(i) for i in shape])
         except TypeError:
-            new_shape = (int(shape), )
-        self._shape = new_shape
-        super(FieldArray, self).__init__()
+            self._shape = (int(shape), )
 
     @property
     def shape(self):

nifty/operators/fft_operator/transformations/gllmtransformation.py

@@ -100,11 +100,6 @@ class GLLMTransformation(SlicingTransformation):
         super(GLLMTransformation, cls).check_codomain(domain, codomain)
 
     def _transformation_of_slice(self, inp, **kwargs):
-        if inp.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(inp.dtype))
-
         nlat = self.domain.nlat
         nlon = self.domain.nlon
         lmax = self.codomain.lmax

nifty/operators/fft_operator/transformations/hplmtransformation.py

@@ -92,11 +92,6 @@ class HPLMTransformation(SlicingTransformation):
         super(HPLMTransformation, cls).check_codomain(domain, codomain)
 
     def _transformation_of_slice(self, inp, **kwargs):
-        if inp.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(inp.dtype))
-
         lmax = self.codomain.lmax
         mmax = lmax
 

nifty/operators/fft_operator/transformations/lmgltransformation.py

@@ -106,11 +106,6 @@ class LMGLTransformation(SlicingTransformation):
         super(LMGLTransformation, cls).check_codomain(domain, codomain)
 
     def _transformation_of_slice(self, inp, **kwargs):
-        if inp.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(inp.dtype))
-
         nlat = self.codomain.nlat
         nlon = self.codomain.nlon
         lmax = self.domain.lmax

nifty/operators/fft_operator/transformations/lmhptransformation.py

@@ -94,11 +94,6 @@ class LMHPTransformation(SlicingTransformation):
         super(LMHPTransformation, cls).check_codomain(domain, codomain)
 
     def _transformation_of_slice(self, inp, **kwargs):
-        if inp.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(inp.dtype))
-
         nside = self.codomain.nside
         lmax = self.domain.lmax
         mmax = lmax

nifty/operators/fft_operator/transformations/rg_transforms.py

@@ -433,11 +433,6 @@ class FFTW(Transform):
                 not all(axis in range(len(val.shape)) for axis in axes):
             raise ValueError("Provided axes does not match array shape")
 
-        if val.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(val.dtype))
-
         # If the input is a numpy array we transform it locally
         if not isinstance(val, distributed_data_object):
             # Cast to a np.ndarray
@@ -582,11 +577,6 @@ class NUMPYFFT(Transform):
                 not all(axis in range(len(val.shape)) for axis in axes):
             raise ValueError("Provided axes does not match array shape")
 
-        if val.dtype not in (np.float, np.complex):
-            self.logger.warn("The input array has dtype: %s. The FFT will "
-                             "be performed at double precision." %
-                             str(val.dtype))
-
         return_val = val.copy_empty(global_shape=val.shape,
                                     dtype=np.complex)
 

nifty/operators/invertible_operator_mixin/invertible_operator_mixin.py

@@ -104,6 +104,7 @@ class InvertibleOperatorMixin(object):
                                                 x0=x0)
         return result
 
+    #MR FIXME: why? shouldn't this be equivalent to the adjoint inverse?
     def _inverse_adjoint_times(self, x, spaces):
         raise NotImplementedError(
             "no generic instance method 'inverse_adjoint_times'.")

nifty/operators/propagator_operator/harmonic_propagator_operator.py

@@ -83,7 +83,7 @@ class HarmonicPropagatorOperator(InvertibleOperatorMixin, EndomorphicOperator):
 
     # ---Overwritten properties and methods---
 
-    def __init__(self, S=None, M=None, R=None, N=None, inverter=None,
+    def __init__(self, S, M=None, R=None, N=None, inverter=None,
                  preconditioner=None):
         """
             Sets the standard operator properties and `codomain`, `_A1`, `_A2`,
@@ -102,7 +102,6 @@ class HarmonicPropagatorOperator(InvertibleOperatorMixin, EndomorphicOperator):
 
         """
         # infer domain, and target
-        # infer domain, and target
         if M is not None:
             self._codomain = M.domain
             self._likelihood = M.times

nifty/operators/smoothing_operator/direct_smoothing_operator.py

@@ -144,6 +144,7 @@ class DirectSmoothingOperator(SmoothingOperator):
             distribution_strategy='not')
         distance_array = distance_array.get_local_data(copy=False)
 
+        #MR FIXME: this causes calls of log(0.) which should probably be avoided
         if self.log_distances:
             np.log(distance_array, out=distance_array)
 

nifty/operators/smoothing_operator/fft_smoothing_operator.py

@@ -24,6 +24,7 @@ class FFTSmoothingOperator(SmoothingOperator):
         kernel = codomain.get_distance_array(
             distribution_strategy=axes_local_distribution_strategy)
 
+        #MR FIXME: this causes calls of log(0.) which should probably be avoided
         if self.log_distances:
             kernel.apply_scalar_function(np.log, inplace=True)
 
@@ -43,6 +44,7 @@ class FFTSmoothingOperator(SmoothingOperator):
 
         # apply the kernel
         if inverse:
+            #MR FIXME: danger of having division by zero or overflows
             local_transformed_x /= local_kernel
         else:
             local_transformed_x *= local_kernel

nifty/spaces/power_space/power_space.py

@@ -261,6 +261,7 @@ class PowerSpace(Space):
         hdf5_group.attrs['nbin'] = str(self.config["nbin"])
         hdf5_group.attrs['binbounds'] = str(self.config["binbounds"])
 
+        #MR FIXME: why not "return None" as happens everywhere else?
         return {
             'harmonic_partner': self.harmonic_partner,
             'pindex': self.pindex,

nifty/spaces/space/space.py

@@ -202,6 +202,4 @@ class Space(DomainObject):
         raise NotImplementedError
 
     def __repr__(self):
-        string = ""
-        string += str(type(self)) + "\n"
-        return string
+        return str(type(self)) + "\n"

nifty/sugar.py

@@ -18,25 +18,22 @@
 
 from nifty import PowerSpace,\
                   Field,\
-                  DiagonalOperator,\
-                  FFTOperator
+                  DiagonalOperator
 
 __all__ = ['create_power_operator']
 
 
 def create_power_operator(domain, power_spectrum, dtype=None,
                           distribution_strategy='not'):
+<<<<<<< nifty/sugar.py
     """ Creates a diagonal operator with the given power spectrum.
 
-    Constructs a diagonal operator that lives over the specified domain, or
-    its default harmonic codomain in case it is not harmonic.
+    Constructs a diagonal operator that lives over the specified domain.
 
     Parameters
     ----------
     domain : DomainObject
-        Domain over which the power operator shall live. If this is not a
-        harmonic domain, it will return an operator for its harmonic codomain
-        instead.
+        Domain over which the power operator shall live. 
     power_spectrum : (array-like, method)
         An array-like object, or a method that implements the square root
         of a power spectrum as a function of k.
@@ -53,21 +50,14 @@ def create_power_operator(domain, power_spectrum, dtype=None,
     DiagonalOperator : An operator that implements the given power spectrum.
 
     """
-
-    if not domain.harmonic:
-        fft = FFTOperator(domain)
-        domain = fft.target[0]
+=======
+>>>>>>> nifty/sugar.py
 
     power_domain = PowerSpace(domain,
                               distribution_strategy=distribution_strategy)
-
-    fp = Field(power_domain,
-               val=power_spectrum, dtype=dtype,
+    fp = Field(power_domain, val=power_spectrum, dtype=dtype,
                distribution_strategy=distribution_strategy)
     fp **= 2
-
     f = fp.power_synthesize(mean=1, std=0, real_signal=False)
 
-    power_operator = DiagonalOperator(domain, diagonal=f, bare=True)
-
-    return power_operator
+    return DiagonalOperator(domain, diagonal=f, bare=True)

test/test_operators/test_diagonal_operator.py

@@ -106,6 +106,7 @@ class DiagonalOperator_Tests(unittest.TestCase):
         assert_allclose(trace_op, np.sum(1./diag.val.get_full_data()))
 
     @expand(product(spaces, [True, False]))
+    #MR FIXME: what if any diagonal element <=0?
     def test_trace_log(self, space, copy):
         diag = Field.from_random('normal', domain=space)
         D = DiagonalOperator(space, diagonal=diag, copy=copy)
@@ -127,8 +128,9 @@ class DiagonalOperator_Tests(unittest.TestCase):
         assert_allclose(inv_det, 1./D.determinant())
 
     @expand(product(spaces, [True, False], [True, False]))
+    #MR FIXME: what if determinant <=0?
     def test_log_determinant(self, space, bare, copy):
         diag = Field.from_random('normal', domain=space)
         D = DiagonalOperator(space, diagonal=diag, bare=bare, copy=copy)
         log_det = D.log_determinant()
-        assert_allclose(log_det, np.log(D.determinant()))
\ No newline at end of file
+        assert_allclose(log_det, np.log(D.determinant()))

test/test_operators/test_smoothing_operator.py

@@ -36,9 +36,6 @@ def _get_rtol(tp):
     else:
         return 1e-5
 
-from itertools import product
-from test.common import expand
-
 class SmoothingOperator_Tests(unittest.TestCase):
     spaces = [RGSpace(100)]