Merge pull request #34 from skjerns/add_sparse_csc

add sparse matrix support

.github/workflows/build_distribute.yml

@@ -24,7 +24,7 @@ jobs:
         pip install setuptools wheel twine
     - name: Run tests once
       run: |
-        pip install flake8 pytest
+        pip install flake8 pytest scipy
         pip install -e .
         pip install numpy h5py
     - name: Test with pytest

.github/workflows/python-package.yml

@@ -27,7 +27,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install flake8 pytest
+        pip install flake8 pytest scipy
         pip install -e .
         pip install numpy h5py
     - name: Lint with flake8

mat73/__init__.py

@@ -122,7 +122,31 @@ def unpack_mat(self, hdf5, depth=0, MATLAB_class=None):
                         MATLAB_class = MATLAB_class.decode()
                 unpacked = self.unpack_mat(elem, depth=depth+1, 
                                            MATLAB_class=MATLAB_class)
-                if MATLAB_class=='struct' and len(elem)>1 and \
+                # sparse elements need to be loaded separately
+                # see https://github.com/skjerns/mat7.3/issues/28
+                if 'MATLAB_sparse' in elem.attrs:
+                    try:
+                        from scipy.sparse import csc_matrix
+                        data = unpacked['data']
+                        row_ind = unpacked['ir']
+                        col_ind = unpacked['jc']                    
+                        n_rows = elem.attrs['MATLAB_sparse']
+                        n_cols = len(col_ind) - 1
+                        unpacked = csc_matrix((data, row_ind, col_ind), shape=(n_rows, n_cols))
+                    except ModuleNotFoundError:
+                        logging.error(f'`scipy` not installed. To load the sparse matrix'
+                                      f' `{elem.name}`,'
+                                      ' you need to have scipy installed. Please install'
+                                      ' via `pip install scipy`')
+                    except DeprecationWarning:
+                        logging.error(f'Tried loading the sparse matrix `{elem.name}`'
+                                      ' with scipy, but'
+                                      ' the interface has been deprecated. Please'
+                                      ' raise this error as an issue on GitHub:'
+                                      ' https://github.com/skjerns/mat7.3/issues')
+
+
+                elif MATLAB_class=='struct' and len(elem)>1 and \
                 isinstance(unpacked, dict):
                     values = unpacked.values()
                     # we can only pack them together in MATLAB style if
@@ -184,7 +208,6 @@ def convert_mat(self, dataset, depth, MATLAB_class=None):
                 logging.error(message)
             return None
 
-        # if has
         known_cls = ['cell', 'char', 'bool', 'logical', 'double', 'single', 
                      'int8', 'int16', 'int32', 'int64', 'uint8', 'uint16',
                      'uint32', 'uint64']
@@ -303,8 +326,9 @@ def savemat(filename, verbose=True):
 
 
 if __name__=='__main__':
-    # d = loadmat('../tests/testfile5.mat', use_attrdict=True)
+    # for testing / debugging
+    d = loadmat('../tests/testfile1.mat')
 
 
-    file = '../tests/testfile8.mat'
-    data = loadmat(file)
+    # file = '../tests/testfile8.mat'
+    # data = loadmat(file)

mat73/version.py

@@ -1 +1 @@
-__version__ = '0.58'
+__version__ = '0.59'

setup.py

@@ -18,6 +18,7 @@
      url="https://github.com/skjerns/mat7.3",
      download_url=f"https://github.com/skjerns/mat7.3/archive/v{__version__}.tar.gz",
      install_requires=['h5py', 'numpy'],
+     tests_require=['scipy'],
      license='MIT',
      packages=['mat73'],
      classifiers=[

tests/create_mat.m

@@ -31,6 +31,7 @@
 data.struct_ = struct('test', [1,2,3,4])
 data.struct2_ = struct('type',{'big','little'},'color','red','x',{single([1.1,1.2,0.3;2,3,4]), double([1.1,1.2,0.3])})
 data.structarr_ = struct('f1', {'some text'; [10,20,30]; magic(5)}, 'f2', {'v1'; 'v2'; 'v3';});
+data.sparse_ = sparse([2, 4], [5, 8], [6, 7], 10, 8);
 
 secondvar = [1,2,3,4]
 

tests/test_mat73.py

@@ -26,6 +26,8 @@
     name = 'unknown'
     message = 'no msg'
 
+EXPECTED_VARS_FILE1 = 30
+
 print(f'#### Installed version: mat73-{version} on {branch}({name}) "{message}" ####')
 
 class Testing(unittest.TestCase):
@@ -51,7 +53,7 @@ def test_file1_noattr(self):
 
         assert len(d)==3
         assert len(d.keys())==3
-        assert len(data)==29
+        assert len(data)==EXPECTED_VARS_FILE1
         assert data['arr_two_three'].shape==(3,2)
         np.testing.assert_allclose(d['secondvar'], [1,2,3,4])
         np.testing.assert_array_equal(data['arr_bool'], np.array([True,True,False]))
@@ -108,6 +110,11 @@ def test_file1_noattr(self):
         assert data['structarr_'][1]['f2'] == ['v2']
         assert data['structarr_'][2]['f2'] == ['v3']
         assert d['keys'] == 'must_not_overwrite'
+        sparse = data['sparse_'].toarray()
+        assert sparse[1,4] == 6
+        assert sparse[3,7] == 7
+        assert (sparse!=0).sum()==2
+
 
         with self.assertRaises(AttributeError):
             d.structarr_
@@ -122,7 +129,7 @@ def test_file1_withattr(self):
 
         assert len(d)==3
         assert len(d.keys())==3
-        assert len(data)==29
+        assert len(data)==EXPECTED_VARS_FILE1
         assert data['arr_two_three'].shape==(3,2)
         np.testing.assert_allclose(d['secondvar'], [1,2,3,4])
         np.testing.assert_array_equal(data['arr_bool'], np.array([True,True,False]))
@@ -183,7 +190,7 @@ def test_file1_withattr(self):
         data = d.data
 
         assert len(d)==3
-        assert len(data)==29
+        assert len(data)==EXPECTED_VARS_FILE1
         assert data.arr_two_three.shape==(3,2)
         np.testing.assert_allclose(d.secondvar, [1,2,3,4])
         np.testing.assert_array_equal(data.arr_bool, np.array([True,True,False]))
@@ -358,7 +365,7 @@ def test_load_specific_vars(self):
 
         data = mat73.loadmat(self.testfile1, only_include=['data', 'keys'])
         assert len(data)==2
-        assert len(data['data'])==29
+        assert len(data['data'])==EXPECTED_VARS_FILE1
         assert len(data['data']['cell_'])==7
 
         # check if loading times are faster, should be the case.