Reduce SCALE memory usage

nimare/meta/cbma/ale.py

@@ -541,18 +541,23 @@ def _scale_to_p(self, stat_values, scale_values):
         -----
         This method also uses the "histogram_bins" element in the null_distributions_ attribute.
         """
-        scale_zeros = scale_values == 0
-        n_zeros = np.sum(scale_zeros, axis=0)
-        scale_values[scale_values == 0] = np.nan
-        scale_hists = np.zeros(
-            ((len(self.null_distributions_["histogram_bins"]),) + n_zeros.shape)
-        )
-        scale_hists[0, :] = n_zeros
-        scale_hists[1:, :] = np.apply_along_axis(self._make_hist, 0, scale_values)
+        p_values = np.empty_like(stat_values)
+
+        for i_voxel in range(stat_values.shape[0]):
+            voxel_null = scale_values[:, i_voxel]
+            scale_zeros = voxel_null == 0
+            n_zeros = np.sum(scale_zeros)
+            voxel_null[scale_zeros] = np.nan
+            scale_hist = np.empty(len(self.null_distributions_["histogram_bins"]))
+            scale_hist[0] = n_zeros
+            scale_hist[1:] = self._make_hist(voxel_null)
+
+            p_values[i_voxel] = nullhist_to_p(
+                stat_values[i_voxel],
+                scale_hist,
+                self.null_distributions_["histogram_bins"],
+            )
 
-        p_values = nullhist_to_p(
-            stat_values, scale_hists, self.null_distributions_["histogram_bins"]
-        )
         z_values = p_to_z(p_values, tail="one")
         return p_values, z_values
 

nimare/meta/utils.py

@@ -450,7 +450,7 @@ def get_ale_kernel(img, sample_size=None, fwhm=None):
         uncertain_subjects = (11.6 / (2 * np.sqrt(2 / np.pi)) * np.sqrt(8 * np.log(2))) / np.sqrt(
             sample_size
         )  # pylint: disable=no-member
-        fwhm = np.sqrt(uncertain_subjects ** 2 + uncertain_templates ** 2)
+        fwhm = np.sqrt(uncertain_subjects**2 + uncertain_templates**2)
 
     fwhm_vox = fwhm / np.sqrt(np.prod(img.header.get_zooms()))
     sigma_vox = (

nimare/stats.py

@@ -103,7 +103,7 @@ def pearson(x, y):
     data = np.vstack((x, y))
     ms = data.mean(axis=1)[(slice(None, None, None), None)]
     datam = data - ms
-    datass = np.sqrt(np.sum(datam ** 2, axis=1))
+    datass = np.sqrt(np.sum(datam**2, axis=1))
     temp = np.dot(datam[1:], datam[0].T)
     rs = temp / (datass[1:] * datass[0])
     return rs

nimare/transforms.py

@@ -545,7 +545,7 @@ def se_to_varcope(se):
     -----
     Sampling variance is standard error squared.
     """
-    varcope = se ** 2
+    varcope = se**2
     return varcope
 
 

nimare/utils.py

@@ -711,12 +711,12 @@ def memmap_context(self, *args, **kwargs):
 
 
 BYTE = 2
-KILOBYTE = BYTE ** 10
+KILOBYTE = BYTE**10
 BYTE_CONVERSION = {
     "kb": KILOBYTE,
-    "mb": KILOBYTE ** 2,
-    "gb": KILOBYTE ** 3,
-    "tb": KILOBYTE ** 4,
+    "mb": KILOBYTE**2,
+    "gb": KILOBYTE**3,
+    "tb": KILOBYTE**4,
 }
 
 

setup.cfg

@@ -50,7 +50,7 @@ install_requires =
     requests
     scikit-learn
     scipy
-    statsmodels
+    statsmodels<0.13.2  # This version doesn't install properly
     tqdm
 packages = find:
 include_package_data = False