Loftusa patch 3

graspologic/align/seedless_procrustes.py

@@ -4,7 +4,6 @@
 from typing import Optional, Union
 
 import numpy as np
-import ot
 from sklearn.utils import check_array
 
 from graspologic.types import Tuple
@@ -285,8 +284,10 @@ def _optimal_transport(
         cost_matrix = (
             np.linalg.norm((X @ Q).reshape(n, 1, d) - Y.reshape(1, m, d), axis=2) ** 2
         )
+        from ot import sinkhorn
         # Note: the type info on ot.sinkhorn can support a return type of Tuple[np.ndarray, Dict[str, Any]], but only
         # if this function is called with log=True, which we aren't doing.
+        import ot
         P: np.ndarray = ot.sinkhorn(
             a=probability_mass_X,
             b=probability_mass_Y,

graspologic/layouts/auto.py

@@ -8,7 +8,6 @@
 
 import networkx as nx
 import numpy as np
-import umap
 from sklearn.manifold import TSNE
 
 from graspologic.types import Dict, List, Tuple
@@ -198,9 +197,9 @@ def layout_umap(
         on Neighbor Embeddings along the Attraction-Repulsion Spectrum. ArXiv e-prints
         2007.08902v1, 17 Jul 2020.
     """
-
+    from umap import UMAP
     lcc_graph, tensors, labels = _node2vec_for_layout(graph, max_edges, random_seed)
-    points = umap.UMAP(
+    points = UMAP(
         min_dist=min_dist, n_neighbors=n_neighbors, random_state=random_seed
     ).fit_transform(tensors)
     positions = _node_positions_from(

graspologic/match/solver.py

@@ -8,7 +8,6 @@
 
 import numpy as np
 from beartype import beartype
-from ot import sinkhorn
 from scipy.optimize import linear_sum_assignment
 from sklearn.utils import check_scalar
 
@@ -363,6 +362,7 @@ def linear_sum_transport(
         self,
         P: np.ndarray,
     ) -> np.ndarray:
+        from ot import sinkhorn
         maximize = self.maximize
         reg = self.transport_regularizer