Merge pull request #18 from cics-nd/optimize

Replace .compute_loss() with .loss()

Author: sdatkinson

gptorch/models/base.py

@@ -236,7 +236,7 @@ def optimize(self, method='Adam', max_iter=2000, verbose=True,
                 for idx in range(max_iter):
                     self.optimizer.zero_grad()
                     # forward
-                    loss = self.compute_loss()
+                    loss = self.loss()
                     # backward
                     loss.backward()
                     self.optimizer.step()
@@ -248,7 +248,7 @@ def optimize(self, method='Adam', max_iter=2000, verbose=True,
 
                     def closure():
                         self.optimizer.zero_grad()
-                        loss = self.compute_loss()
+                        loss = self.loss()
                         loss.backward()
                         return loss
 
@@ -260,7 +260,7 @@ def closure():
                 for idx in range(max_iter):
                     self.optimizer.zero_grad()
                     # forward
-                    loss = self.compute_loss()
+                    loss = self.loss()
                     # backward
                     loss.backward()
                     self.optimizer.step()
@@ -272,7 +272,7 @@ def closure():
 
                     def closure():
                         self.optimizer.zero_grad()
-                        loss = self.compute_loss()
+                        loss = self.loss()
                         loss.backward()
                         return loss
 

setup.py

@@ -24,7 +24,7 @@
 ]
 
 setup(name="gptorch",
-    version="0.3.0",
+    version="0.3.1",
     description="gptorch - a Gaussian process toolbox built on PyTorch",
     author="Yinhao Zhu, Steven Atkinson",
     author_email="yzhu10@nd.edu, satkinso@nd.edu",

test/test_models/test_base.py

@@ -47,6 +47,11 @@ def test_cpu(self):
         assert not gp.X.is_cuda
         assert not gp.Y.is_cuda
 
+    def test_optimize(self):
+        gp = self._get_model()
+        gp.optimize(max_iter=2)
+        gp.optimize(method="L-BFGS-B", max_iter=2)
+
     def test_predict_f(self):
         self._predict_fy("predict_f")