[MRG] FIX Fix LMNN rollback

metric_learn/lmnn.py

@@ -90,83 +90,49 @@ def fit(self, X, y):
       a1[nn_idx] = np.array([])
       a2[nn_idx] = np.array([])
 
-    # initialize gradient and L
-    G = dfG * reg + df * (1-reg)
+    # initialize L
     L = self.L_
-    objective = np.inf
-
-    # main loop
-    for it in xrange(1, self.max_iter):
-      df_old = df.copy()
-      a1_old = [a.copy() for a in a1]
-      a2_old = [a.copy() for a in a2]
-      objective_old = objective
-      # Compute pairwise distances under current metric
-      Lx = L.dot(self.X_.T).T
-      g0 = _inplace_paired_L2(*Lx[impostors])
-      Ni = 1 + _inplace_paired_L2(Lx[target_neighbors], Lx[:,None,:])
-      g1,g2 = Ni[impostors]
-
-      # compute the gradient
-      total_active = 0
-      for nn_idx in reversed(xrange(k)):
-        act1 = g0 < g1[:,nn_idx]
-        act2 = g0 < g2[:,nn_idx]
-        total_active += act1.sum() + act2.sum()
-
-        if it > 1:
-          plus1 = act1 & ~a1[nn_idx]
-          minus1 = a1[nn_idx] & ~act1
-          plus2 = act2 & ~a2[nn_idx]
-          minus2 = a2[nn_idx] & ~act2
-        else:
-          plus1 = act1
-          plus2 = act2
-          minus1 = np.zeros(0, dtype=int)
-          minus2 = np.zeros(0, dtype=int)
-
-        targets = target_neighbors[:,nn_idx]
-        PLUS, pweight = _count_edges(plus1, plus2, impostors, targets)
-        df += _sum_outer_products(self.X_, PLUS[:,0], PLUS[:,1], pweight)
-        MINUS, mweight = _count_edges(minus1, minus2, impostors, targets)
-        df -= _sum_outer_products(self.X_, MINUS[:,0], MINUS[:,1], mweight)
-
-        in_imp, out_imp = impostors
-        df += _sum_outer_products(self.X_, in_imp[minus1], out_imp[minus1])
-        df += _sum_outer_products(self.X_, in_imp[minus2], out_imp[minus2])
-
-        df -= _sum_outer_products(self.X_, in_imp[plus1], out_imp[plus1])
-        df -= _sum_outer_products(self.X_, in_imp[plus2], out_imp[plus2])
-
-        a1[nn_idx] = act1
-        a2[nn_idx] = act2
-
-      # do the gradient update
-      assert not np.isnan(df).any()
-      G = dfG * reg + df * (1-reg)
 
-      # compute the objective function
-      objective = total_active * (1-reg)
-      objective += G.flatten().dot(L.T.dot(L).flatten())
-      assert not np.isnan(objective)
-      delta_obj = objective - objective_old
+    # first iteration: we compute variables (including objective and gradient)
+    #  at initialization point
+    G, objective, total_active, df, a1, a2 = (
+        self._loss_grad(L, dfG, impostors, 1, k, reg, target_neighbors, df, a1,
+                        a2))
+
+    for it in xrange(2, self.max_iter):
+      # then at each iteration, we try to find a value of L that has better
+      # objective than the previous L, following the gradient:
+      while True:
+        # the next point next_L to try out is found by a gradient step
+        L_next = L - 2 * learn_rate * G
+        # we compute the objective at next point
+        # we copy variables that can be modified by _loss_grad, because if we
+        # retry we don t want to modify them several times
+        (G_next, objective_next, total_active_next, df_next, a1_next,
+         a2_next) = (
+            self._loss_grad(L_next, dfG, impostors, it, k, reg,
+                            target_neighbors, df.copy(), list(a1), list(a2)))
+        assert not np.isnan(objective)
+        delta_obj = objective_next - objective
+        if delta_obj > 0:
+          # if we did not find a better objective, we retry with an L closer to
+          # the starting point, by decreasing the learning rate (making the
+          # gradient step smaller)
+          learn_rate /= 2
+        else:
+          # otherwise, if we indeed found a better obj, we get out of the loop
+          break
+      # when the better L is found (and the related variables), we set the
+      # old variables to these new ones before next iteration and we
+      # slightly increase the learning rate
+      L = L_next
+      G, df, objective, total_active, a1, a2 = (
+          G_next, df_next, objective_next, total_active_next, a1_next, a2_next)
+      learn_rate *= 1.01
 
       if self.verbose:
         print(it, objective, delta_obj, total_active, learn_rate)
 
-      # update step size
-      if delta_obj > 0:
-        # we're getting worse... roll back!
-        learn_rate /= 2.0
-        df = df_old
-        a1 = a1_old
-        a2 = a2_old
-        objective = objective_old
-      else:
-        # update L
-        L -= learn_rate * 2 * L.dot(G)
-        learn_rate *= 1.01
-
       # check for convergence
       if it > self.min_iter and abs(delta_obj) < self.convergence_tol:
         if self.verbose:
@@ -181,6 +147,54 @@ def fit(self, X, y):
     self.n_iter_ = it
     return self
 
+  def _loss_grad(self, L, dfG, impostors, it, k, reg, target_neighbors, df, a1,
+                 a2):
+    # Compute pairwise distances under current metric
+    Lx = L.dot(self.X_.T).T
+    g0 = _inplace_paired_L2(*Lx[impostors])
+    Ni = 1 + _inplace_paired_L2(Lx[target_neighbors], Lx[:, None, :])
+    g1, g2 = Ni[impostors]
+    # compute the gradient
+    total_active = 0
+    for nn_idx in reversed(xrange(k)):
+      act1 = g0 < g1[:, nn_idx]
+      act2 = g0 < g2[:, nn_idx]
+      total_active += act1.sum() + act2.sum()
+
+      if it > 1:
+        plus1 = act1 & ~a1[nn_idx]
+        minus1 = a1[nn_idx] & ~act1
+        plus2 = act2 & ~a2[nn_idx]
+        minus2 = a2[nn_idx] & ~act2
+      else:
+        plus1 = act1
+        plus2 = act2
+        minus1 = np.zeros(0, dtype=int)
+        minus2 = np.zeros(0, dtype=int)
+
+      targets = target_neighbors[:, nn_idx]
+      PLUS, pweight = _count_edges(plus1, plus2, impostors, targets)
+      df += _sum_outer_products(self.X_, PLUS[:, 0], PLUS[:, 1], pweight)
+      MINUS, mweight = _count_edges(minus1, minus2, impostors, targets)
+      df -= _sum_outer_products(self.X_, MINUS[:, 0], MINUS[:, 1], mweight)
+
+      in_imp, out_imp = impostors
+      df += _sum_outer_products(self.X_, in_imp[minus1], out_imp[minus1])
+      df += _sum_outer_products(self.X_, in_imp[minus2], out_imp[minus2])
+
+      df -= _sum_outer_products(self.X_, in_imp[plus1], out_imp[plus1])
+      df -= _sum_outer_products(self.X_, in_imp[plus2], out_imp[plus2])
+
+      a1[nn_idx] = act1
+      a2[nn_idx] = act2
+    # do the gradient update
+    assert not np.isnan(df).any()
+    G = dfG * reg + df * (1 - reg)
+    # compute the objective function
+    objective = total_active * (1 - reg)
+    objective += G.flatten().dot(L.T.dot(L).flatten())
+    return G, objective, total_active, df, a1, a2
+
   def _select_targets(self):
     target_neighbors = np.empty((self.X_.shape[0], self.k), dtype=int)
     for label in self.labels_:

test/metric_learn_test.py

@@ -1,5 +1,5 @@
-import re
 import unittest
+import re
 import pytest
 import numpy as np
 from scipy.optimize import check_grad
@@ -76,6 +76,37 @@ def test_iris(self):
       self.assertLess(csep, 0.25)
 
 
+def test_convergence_simple_example(capsys):
+  # LMNN should converge on this simple example, which it did not with
+  # this issue: https://github.com/metric-learn/metric-learn/issues/88
+  X, y = make_classification(random_state=0)
+  lmnn = python_LMNN(verbose=True)
+  lmnn.fit(X, y)
+  out, _ = capsys.readouterr()
+  assert "LMNN converged with objective" in out
+
+
+def test_no_twice_same_objective(capsys):
+  # test that the objective function never has twice the same value
+  # see https://github.com/metric-learn/metric-learn/issues/88
+  X, y = make_classification(random_state=0)
+  lmnn = python_LMNN(verbose=True)
+  lmnn.fit(X, y)
+  out, _ = capsys.readouterr()
+  lines = re.split("\n+", out)
+  # we get only objectives from each line:
+  # the regexp matches a float that follows an integer (the iteration
+  # number), and which is followed by a (signed) float (delta obj). It
+  # matches for instance:
+  # 3 **1113.7665747189938** -3.182774197440267 46431.0200999999999998e-06
+  objectives = [re.search("\d* (?:(\d*.\d*))[ | -]\d*.\d*", s)
+                for s in lines]
+  objectives = [match.group(1) for match in objectives if match is not None]
+  # we remove the last element because it can be equal to the penultimate
+  # if the last gradient update is null
+  assert len(objectives[:-1]) == len(set(objectives[:-1]))
+
+
 class TestSDML(MetricTestCase):
   def test_iris(self):
     # Note: this is a flaky test, which fails for certain seeds.