RF/OPT boosting: use Cython mp (#50)

.gitignore

@@ -42,3 +42,5 @@ doc/auto_examples
 *.ipynb
 .ipynb_checkpoints
 .eggs
+
+*.html

doc/changes.rst

@@ -3,6 +3,12 @@ Changes
 
 .. currentmodule:: eelbrain
 
+New in 0.38
+-----------
+
+* :func:`boosting` optimized (as a consequence, the progress bar has been disabled).
+
+
 New in 0.37
 -----------
 

eelbrain/_data_obj.py

@@ -1266,7 +1266,11 @@ def names(self):
 
 class Named:
 
-    def __init__(self, name, info):
+    def __init__(
+            self,
+            name: str = None,
+            info: Dict = None,
+    ):
         self.info = {} if info is None else dict(info)
         self._name = name
 

eelbrain/_ndvar.py

@@ -15,7 +15,7 @@
 from typing import Any, Callable, Literal, Sequence, Union
 
 import mne
-from numba import njit, prange
+import numba
 import numpy as np
 import numpy
 from scipy import linalg, ndimage, signal, stats
@@ -28,7 +28,7 @@
 from ._mne import complete_source_space
 from ._stats.connectivity import Connectivity
 from ._stats.connectivity import find_peaks as _find_peaks
-from ._trf._boosting_opt import l1
+from ._trf._fit_metrics import error_for_indexes
 from ._utils.numpy_utils import aslice, newaxis
 
 
@@ -272,7 +272,7 @@ def convolve(h, x, ds=None, name=None):
     return NDVar(out, dims, *op_name(x, name=name))
 
 
-@njit(parallel=True)
+@numba.njit(nogil=True, cache=True, parallel=True)
 def parallel_convolve(
         h_flat: np.ndarray,  # n_h_only, n_shared, n_h_times
         x_flat: np.ndarray,  # n_x_only, n_shared, n_x_times
@@ -282,12 +282,12 @@ def parallel_convolve(
 ):
     # loop through x and h dimensions
     out_indexes = [(ix, ih) for ix in range(len(x_flat)) for ih in range(len(h_flat))]
-    for i_out in prange(len(out_indexes)):
+    for i_out in numba.prange(len(out_indexes)):
         i_x, i_h = out_indexes[i_out]
         convolve_jit(h_flat[i_h], x_flat[i_x], out_flat[i_x, i_h], i_start, i_stop)
 
 
-@njit
+@numba.njit(nogil=True, cache=True)
 def convolve_jit(
         h: np.ndarray,  # n_h, n_h_times
         x: np.ndarray,  # n_h, n_x_times
@@ -805,7 +805,7 @@ def label_operator(labels, operation='mean', exclude=None, weights=None,
         if weights is not None:
             xs *= weights
         if operation == 'mean':
-            xs /= l1(xs, np.array(((0, len(xs)),), np.int64))
+            xs /= error_for_indexes(xs, np.array(((0, len(xs)),), np.int64), 1)
     return NDVar(x, (label_dim, dim), labels.name)