Fix deprecation of shape assignment in numpy 2.5+

nitime/_compat.py

@@ -1,5 +1,46 @@
+"""Compatibility utilities for different dependency versions."""
+
+import numpy as np
+from packaging.version import Version
+
+
+def _reshape_view(arr, shape):
+    """Reshape an array as a view, raising if a copy would be required.
+
+    This function provides compatibility across NumPy versions for reshaping
+    arrays as views. On NumPy >= 2.1, it uses ``reshape(copy=False)`` which
+    explicitly fails if a view cannot be created. On older versions, it uses
+    direct shape assignment which has the same behavior but is deprecated in
+    NumPy 2.5+.
+
+    Parameters
+    ----------
+    arr : ndarray
+        The array to reshape.
+    shape : tuple of int
+        The new shape.
+
+    Returns
+    -------
+    ndarray
+        A reshaped view of the array.
+
+    Raises
+    ------
+    AttributeError
+        If a view cannot be created on NumPy < 2.1.
+    ValueError
+        If a view cannot be created on NumPy >= 2.1.
+    """
+    if Version(np.__version__) >= Version("2.1"):
+        return arr.reshape(shape, copy=False)
+    else:
+        arr.shape = shape
+        return arr
+
+
 # np.trapezoid was introduced and np.trapz deprecated in numpy 2.0
-try:  # NP2
+if Version(np.__version__) >= Version("2.0"):
     from numpy import trapezoid
-except ImportError:  # NP1
+else:
     from numpy import trapz as trapezoid

nitime/algorithms/spectral.py

@@ -20,6 +20,8 @@
 # To support older versions of numpy that don't have tril_indices:
 from nitime.index_utils import tril_indices, triu_indices
 
+from nitime._compat import _reshape_view
+
 
 # Set global variables for the default NFFT to be used in spectral analysis and
 # the overlap:
@@ -308,7 +310,7 @@ def periodogram_csd(s, Fs=2 * np.pi, Sk=None, NFFT=None, sides='default',
 
     """
     s_shape = s.shape
-    s.shape = (-1, s_shape[-1])
+    s = _reshape_view(s, (-1, s_shape[-1]))
     # defining an Sk_loc is a little opaque, but it avoids having to
     # reset the shape of any user-given Sk later on
     if Sk is not None:
@@ -322,7 +324,7 @@ def periodogram_csd(s, Fs=2 * np.pi, Sk=None, NFFT=None, sides='default',
             N = s.shape[-1]
         Sk_loc = fftpack.fft(s, n=N)
     # reset s.shape
-    s.shape = s_shape
+    s = _reshape_view(s, s_shape)
 
     M = Sk_loc.shape[0]
 
@@ -550,7 +552,7 @@ def multi_taper_psd(
     NFFT = spectra.shape[-1]
     K = len(eigvals)
     # collapse spectra's shape back down to 3 dimensions
-    spectra.shape = (M, K, NFFT)
+    spectra = _reshape_view(spectra, (M, K, NFFT))
 
     last_freq = NFFT // 2 + 1 if sides == 'onesided' else NFFT
 
@@ -593,12 +595,12 @@ def multi_taper_psd(
         freqs = np.linspace(0, Fs, NFFT, endpoint=False)
 
     out_shape = s.shape[:-1] + (len(freqs),)
-    sdf_est.shape = out_shape
+    sdf_est = _reshape_view(sdf_est, out_shape)
     if jackknife:
-        jk_var.shape = out_shape
+        jk_var = _reshape_view(jk_var, out_shape)
         return freqs, sdf_est, jk_var
     else:
-        nu.shape = out_shape
+        nu = _reshape_view(nu, out_shape)
         return freqs, sdf_est, nu
 
 
@@ -690,7 +692,7 @@ def multi_taper_csd(s, Fs=2 * np.pi, NW=None, BW=None, low_bias=True,
     NFFT = spectra.shape[-1]
     K = len(eigvals)
     # collapse spectra's shape back down to 3 dimensions
-    spectra.shape = (M, K, NFFT)
+    spectra = _reshape_view(spectra, (M, K, NFFT))
 
     # compute the cross-spectral density functions
     last_freq = NFFT // 2 + 1 if sides == 'onesided' else NFFT

nitime/tests/test_timeseries.py

@@ -922,7 +922,7 @@ def test_timearray_math_functions(f, tu):
     "Calling TimeArray.min() .max(), mean() should return TimeArrays"
     a = np.arange(2, 11)
     b = ts.TimeArray(a, time_unit=tu)
-    if f == "ptp" and ts._NP_2:
+    if f == "ptp":
         want = np.ptp(a)
     else:
         want = getattr(a, f)()

nitime/timeseries.py

@@ -33,10 +33,6 @@
 # Our own
 from nitime import descriptors as desc
 
-try:
-    _NP_2 = int(np.__version__.split(".")[0]) >= 2
-except Exception:
-    _NP_2 = True
 
 #-----------------------------------------------------------------------------
 # Module globals
@@ -319,11 +315,7 @@ def mean(self, *args, **kwargs):
         return ret
 
     def ptp(self, *args, **kwargs):
-        if _NP_2:
-            ptp = np.ptp
-        else:
-            ptp = np.ndarray.ptp
-        ret = TimeArray(ptp(self, *args, **kwargs),
+        ret = TimeArray(np.ptp(self, *args, **kwargs),
                         time_unit=base_unit)
         ret.convert_unit(self.time_unit)
         return ret

nitime/utils.py

@@ -13,6 +13,8 @@
 from nitime.lazy import scipy_stats_distributions as dists
 from nitime.lazy import scipy_interpolate as interpolate
 
+from nitime._compat import _reshape_view
+
 
 #-----------------------------------------------------------------------------
 # Spectral estimation testing utilities
@@ -756,7 +758,7 @@ def tapered_spectra(s, tapers, NFFT=None, low_bias=True):
     # compute the y_{i,k}(f) -- full FFT takes ~1.5x longer, but unpacking
     # results of real-valued FFT eats up memory
     t_spectra = fftpack.fft(tapered, n=NFFT, axis=-1)
-    t_spectra.shape = rest_of_dims + (K, NFFT)
+    t_spectra = _reshape_view(t_spectra, rest_of_dims + (K, NFFT))
     if eigvals is None:
         return t_spectra
     return t_spectra, eigvals
@@ -834,7 +836,7 @@ def detect_lines(s, tapers, p=None, **taper_kws):
     numr[...,0] = 1; # don't care about DC
     # denominator -- strength of residual
     spectra = np.rollaxis(spectra, -2, 0)
-    U0.shape = (K,) + (1,) * (spectra.ndim-1)
+    U0 = _reshape_view(U0, (K,) + (1,) * (spectra.ndim-1))
     denomr = spectra - U0*mu
     denomr = np.sum(np.abs(denomr)**2, axis=0) / (2*K-2)
     denomr[...,0] = 1;

pyproject.toml

@@ -33,6 +33,7 @@ classifiers = [
 dependencies = [
     "matplotlib>=3.7",
     "numpy>=1.24",
+    "packaging",
     "scipy>=1.10",
 ]