Implementation of MMC (#61)

* Implementation of PGDM

* Python2 compatibility

* Speed up PGDM on high-dimensional data

* Optimized some summations using `np.einsum`

* Addressed requests from review by perimosocordiae

* Renamed PGDM to MMC

* Addressed 2nd review by perimosocordiae

Author: Callidior

README.rst

@@ -15,6 +15,7 @@ Metric Learning algorithms in Python.
 -  Local Fisher Discriminant Analysis (LFDA)
 -  Relative Components Analysis (RCA)
 -  Metric Learning for Kernel Regression (MLKR)
+-  Mahalanobis Metric for Clustering (MMC)
 
 **Dependencies**
 

metric_learn/__init__.py

@@ -10,3 +10,4 @@
 from .lfda import LFDA
 from .rca import RCA, RCA_Supervised
 from .mlkr import MLKR
+from .mmc import MMC, MMC_Supervised

metric_learn/_util.py

@@ -0,0 +1,12 @@
+import numpy as np
+
+
+# hack around lack of axis kwarg in older numpy versions
+try:
+  np.linalg.norm([[4]], axis=1)
+except TypeError:
+  def vector_norm(X):
+    return np.apply_along_axis(np.linalg.norm, 1, X)
+else:
+  def vector_norm(X):
+    return np.linalg.norm(X, axis=1)

metric_learn/itml.py

@@ -21,6 +21,7 @@
 
 from .base_metric import BaseMetricLearner
 from .constraints import Constraints
+from ._util import vector_norm
 
 
 class ITML(BaseMetricLearner):
@@ -54,10 +55,10 @@ def _process_inputs(self, X, constraints, bounds):
     self.X_ = X = check_array(X)
     # check to make sure that no two constrained vectors are identical
     a,b,c,d = constraints
-    ident = _vector_norm(X[a] - X[b]) > 1e-9
-    a, b = a[ident], b[ident]
-    ident = _vector_norm(X[c] - X[d]) > 1e-9
-    c, d = c[ident], d[ident]
+    no_ident = vector_norm(X[a] - X[b]) > 1e-9
+    a, b = a[no_ident], b[no_ident]
+    no_ident = vector_norm(X[c] - X[d]) > 1e-9
+    c, d = c[no_ident], d[no_ident]
     # init bounds
     if bounds is None:
       self.bounds_ = np.percentile(pairwise_distances(X), (5, 95))
@@ -138,16 +139,6 @@ def fit(self, X, constraints, bounds=None):
   def metric(self):
     return self.A_
 
-# hack around lack of axis kwarg in older numpy versions
-try:
-  np.linalg.norm([[4]], axis=1)
-except TypeError:
-  def _vector_norm(X):
-    return np.apply_along_axis(np.linalg.norm, 1, X)
-else:
-  def _vector_norm(X):
-    return np.linalg.norm(X, axis=1)
-
 
 class ITML_Supervised(ITML):
   """Information Theoretic Metric Learning (ITML)"""