Add an upper limit on the numpy version due to incompatabilities with…

… tslearn. Replaced np.NaN with np.nan for future numpy compatability.

pyspi/base.py

@@ -86,7 +86,7 @@ def multivariate(self,data):
         """ Compute the dependency statistics for the entire multivariate dataset
         """
         A = np.empty((data.n_processes,data.n_processes))
-        A[:] = np.NaN
+        A[:] = np.nan
 
         for j in range(data.n_processes):
             for i in [ii for ii in range(data.n_processes) if ii != j]:

pyspi/calculator.py

@@ -266,7 +266,7 @@ def load_dataset(self, dataset):
         self._table = pd.DataFrame(
             data=np.full(
                 (self.dataset.n_processes, self.n_spis * self.dataset.n_processes),
-                np.NaN,
+                np.nan,
             ),
             columns=columns,
             index=self._dataset.procnames,
@@ -289,13 +289,13 @@ def compute(self):
                 S = self._spis[spi].multivariate(self.dataset)
 
                 # Ensure the diagonal is NaN (sometimes set within the functions)
-                np.fill_diagonal(S, np.NaN)
+                np.fill_diagonal(S, np.nan)
 
                 # Save results
                 self._table[spi] = S
             except Exception as err:
                 warnings.warn(f'Caught {type(err)} for SPI "{spi}": {err}')
-                self._table[spi] = np.NaN
+                self._table[spi] = np.nan
         pbar.close()
         print(f"\nCalculation complete. Time taken: {pbar.format_dict['elapsed']:.4f}s")
         inspect_calc_results(self)

pyspi/statistics/infotheory.py

@@ -297,7 +297,7 @@ def bivariate(self, data, i=None, j=None, verbose=False):
             logging.warning(
                 "MI calcs failed. Maybe check input data for Cholesky factorisation?"
             )
-            return np.NaN
+            return np.nan
 
 
 class TimeLaggedMutualInfo(MutualInfo):
@@ -331,7 +331,7 @@ def bivariate(self, data, i=None, j=None, verbose=False):
             logging.warning(
                 "Time-lagged MI calcs failed. Maybe check input data for Cholesky factorisation?"
             )
-            return np.NaN
+            return np.nan
 
 
 class TransferEntropy(JIDTBase, Directed):
@@ -405,7 +405,7 @@ def bivariate(self, data, i=None, j=None, verbose=False):
             return self._calc.computeAverageLocalOfObservations()
         except Exception as err:
             logging.warning(f"TE calcs failed: {err}.")
-            return np.NaN
+            return np.nan
 
 
 class CrossmapEntropy(JIDTBase, Directed):

requirements.txt

@@ -2,7 +2,7 @@ pytest
 h5py
 scikit-learn
 scipy
-numpy
+numpy<2.0.0
 pandas
 statsmodels
 pyyaml

setup.py

@@ -11,7 +11,7 @@
         'h5py',
         'scikit-learn',
         'scipy',
-        'numpy',
+        'numpy<2.0.0',
         'pandas',
         'statsmodels',
         'pyyaml',

tests/test_calc.py

@@ -98,7 +98,7 @@ def test_pass_dataset_with_nan_into_calculator(nan_loc, expected_output):
     """Check whether ValueError is raised when a dataset containing a NaN is passed into the calculator object"""
     base_dataset = np.random.randn(5, 100)
     for loc in nan_loc:
-        base_dataset[loc, 0] = np.NaN
+        base_dataset[loc, 0] = np.nan
     with pytest.raises(ValueError) as excinfo:
         calc = Calculator(dataset=base_dataset)
     assert f"non-numerics (NaNs) in processes: {expected_output}" in str(excinfo), "NaNs not detected in dataset when loading into Calculator!"