Remove use of .value in mcrbm

Author: lamblin

code/convolutional_mlp.py

@@ -78,10 +78,11 @@ def __init__(self, rng, input, filter_shape, image_shape, poolsize=(2,2)):
         fan_out = filter_shape[0] * numpy.prod(filter_shape[2:]) / numpy.prod(poolsize)
         # replace weight values with random weights
         W_bound = numpy.sqrt(6./(fan_in + fan_out))
-        self.W.value = numpy.asarray( 
+        self.W.set_value(numpy.asarray(
                 rng.uniform(low=-W_bound, high=W_bound, size=filter_shape),
-                dtype = theano.config.floatX)
-  
+                dtype = theano.config.floatX),
+            borrow=True)
+
         # downsample each feature map individually, using maxpooling
         pooled_out = downsample.max_pool_2d( input = conv_out, 
                                     ds = poolsize, ignore_border=True)
@@ -124,9 +125,9 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200, dataset='../data/mnist.pkl.
 
 
     # compute number of minibatches for training, validation and testing
-    n_train_batches = train_set_x.value.shape[0] / batch_size
-    n_valid_batches = valid_set_x.value.shape[0] / batch_size
-    n_test_batches  = test_set_x.value.shape[0]  / batch_size
+    n_train_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
+    n_valid_batches = valid_set_x.get_value(borrow=True).shape[0] / batch_size
+    n_test_batches  = test_set_x.get_value(borrow=True).shape[0]  / batch_size
 
     # allocate symbolic variables for the data
     index = T.lscalar()    # index to a [mini]batch 

code/mcrbm/hmc.py

@@ -392,4 +392,4 @@ def draw(self, **kwargs):
             Numpy matrix whose of dimensions similar to `initial_position`.
        """
         self.simulate()
-        return self.positions.value.copy()
+        return self.positions.get_value(borrow=False)

code/mcrbm/mcRBM.py

@@ -372,7 +372,7 @@ def n_visible_units(self):
         
         """
         try:
-            return self.W.value.shape[0]
+            return self.W.get_value(borrow=True).shape[0]
         except AttributeError:
             return self.W.shape[0]
 
@@ -384,7 +384,7 @@ def n_hidden_cov_units(self):
         
         """
         try:
-            return self.U.value.shape[1]
+            return self.U.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.U.shape[1]
 
@@ -396,7 +396,7 @@ def n_hidden_mean_units(self):
         
         """
         try:
-            return self.W.value.shape[1]
+            return self.W.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.W.shape[1]
 
@@ -545,7 +545,7 @@ def n_hidden_cov_units(self):
         
         """
         try:
-            return self.P.value.shape[1]
+            return self.P.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.P.shape[1]
 
@@ -630,7 +630,7 @@ def alloc_for_P(cls, rbm, visible_batch, batchsize, initial_lr_per_example=0.075
         rval = cls.alloc(rbm, visible_batch, batchsize, initial_lr_per_example, rng, l1_penalty,
                 l1_penalty_start, learn_rate_multipliers, lr_anneal_start, persistent_chains)
 
-        rval.p_mask = sharedX((rbm.P.value!=0).astype('float32'), 'p_mask')
+        rval.p_mask = sharedX((rbm.P.get_value(borrow=True) !=0 ).astype('float32'), 'p_mask')
 
         rval.p_lr = p_lr
         rval.p_training_start=p_training_start

code/mcrbm/mcrbm.py

@@ -372,7 +372,7 @@ def n_visible_units(self):
         
         """
         try:
-            return self.W.value.shape[0]
+            return self.W.get_value(borrow=True).shape[0]
         except AttributeError:
             return self.W.shape[0]
 
@@ -384,7 +384,7 @@ def n_hidden_cov_units(self):
         
         """
         try:
-            return self.U.value.shape[1]
+            return self.U.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.U.shape[1]
 
@@ -396,7 +396,7 @@ def n_hidden_mean_units(self):
         
         """
         try:
-            return self.W.value.shape[1]
+            return self.W.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.W.shape[1]
 
@@ -545,7 +545,7 @@ def n_hidden_cov_units(self):
         
         """
         try:
-            return self.P.value.shape[1]
+            return self.P.get_value(borrow=True).shape[1]
         except AttributeError:
             return self.P.shape[1]
 
@@ -630,7 +630,7 @@ def alloc_for_P(cls, rbm, visible_batch, batchsize, initial_lr_per_example=0.075
         rval = cls.alloc(rbm, visible_batch, batchsize, initial_lr_per_example, rng, l1_penalty,
                 l1_penalty_start, learn_rate_multipliers, lr_anneal_start, persistent_chains)
 
-        rval.p_mask = sharedX((rbm.P.value!=0).astype('float32'), 'p_mask')
+        rval.p_mask = sharedX((rbm.P.get_value(borrow=True) != 0).astype('float32'), 'p_mask')
 
         rval.p_lr = p_lr
         rval.p_training_start=p_training_start

code/mcrbm/test_hmc.py

@@ -41,15 +41,15 @@ def gaussian_energy(x):
     print 'empirical_cov:\n', np.cov(samples.T)
 
     print '****** HMC INTERNALS ******' 
-    print 'final stepsize', sampler.stepsize.value
-    print 'final acceptance_rate', sampler.avg_acceptance_rate.value
+    print 'final stepsize', sampler.stepsize.get_value()
+    print 'final acceptance_rate', sampler.avg_acceptance_rate.get_value()
 
     return sampler
 
 def test_hmc():
     sampler = sampler_on_nd_gaussian(HMC_sampler.new_from_shared_positions,
             burnin=1000, n_samples=1000, dim=5)
     assert abs(sampler.avg_acceptance_rate - sampler.target_acceptance_rate) < .1
-    assert sampler.stepsize.value >= sampler.stepsize_min
-    assert sampler.stepsize.value <= sampler.stepsize_max
+    assert sampler.stepsize.get_value() >= sampler.stepsize_min
+    assert sampler.stepsize.get_value() <= sampler.stepsize_max
 

code/mcrbm/test_mcrbm.py

@@ -82,31 +82,31 @@ def test_reproduce_ranzato_hinton_2010(dataset='MAR',
         if print_jj:
             tile(imgs_fn(jj), "imgs_%06i.png"%jj)
             if persistent_chains:
-                tile(smplr.positions.value, "sample_%06i.png"%jj)
-            tile(rbm.U.value.T, "U_%06i.png"%jj)
-            tile(rbm.W.value.T, "W_%06i.png"%jj)
+                tile(smplr.positions.get_value(borrow=True), "sample_%06i.png"%jj)
+            tile(rbm.U.get_value(borrow=True).T, "U_%06i.png"%jj)
+            tile(rbm.W.get_value(borrow=True).T, "W_%06i.png"%jj)
 
             print 'saving samples', jj, 'epoch', jj/(epoch_size/batchsize)
 
-            print 'l2(U)', l2(rbm.U.value),
-            print 'l2(W)', l2(rbm.W.value),
+            print 'l2(U)', l2(rbm.U.get_value(borrow=True)),
+            print 'l2(W)', l2(rbm.W.get_value(borrow=True)),
             print 'l1_penalty', 
             try:
-                print trainer.effective_l1_penalty.value
+                print trainer.effective_l1_penalty.get_value()
             except:
                 print trainer.effective_l1_penalty
 
-            print 'U min max', rbm.U.value.min(), rbm.U.value.max(),
-            print 'W min max', rbm.W.value.min(), rbm.W.value.max(),
-            print 'a min max', rbm.a.value.min(), rbm.a.value.max(),
-            print 'b min max', rbm.b.value.min(), rbm.b.value.max(),
-            print 'c min max', rbm.c.value.min(), rbm.c.value.max()
+            print 'U min max', rbm.U.get_value(borrow=True).min(), rbm.U.get_value(borrow=True).max(),
+            print 'W min max', rbm.W.get_value(borrow=True).min(), rbm.W.get_value(borrow=True).max(),
+            print 'a min max', rbm.a.get_value(borrow=True).min(), rbm.a.get_value(borrow=True).max(),
+            print 'b min max', rbm.b.get_value(borrow=True).min(), rbm.b.get_value(borrow=True).max(),
+            print 'c min max', rbm.c.get_value(borrow=True).min(), rbm.c.get_value(borrow=True).max()
 
             if persistent_chains:
-                print 'parts min', smplr.positions.value.min(), 
-                print 'max',smplr.positions.value.max(),
-            print 'HMC step', smplr.stepsize.value,
-            print 'arate', smplr.avg_acceptance_rate.value
+                print 'parts min', smplr.positions.get_value(borrow=True).min(),
+                print 'max',smplr.positions.get_value(borrow=True).max(),
+            print 'HMC step', smplr.stepsize.get_value(),
+            print 'arate', smplr.avg_acceptance_rate.get_value()
 
 
         l2_of_Ugrad = learn_fn(jj)

doc/hmc.txt

@@ -583,7 +583,7 @@ elements are:
                 Numpy matrix whose of dimensions similar to `initial_position`.
            """
             self.simulate()
-            return self.positions.value.copy()
+            return self.positions.get_value(borrow=False)
 
 
 Testing our Sampler
@@ -641,17 +641,17 @@ compare the empirical mean and covariance matrix to their true values.
         print 'empirical_cov:\n', np.cov(samples.T)
 
         print '****** HMC INTERNALS ******' 
-        print 'final stepsize', sampler.stepsize.value
-        print 'final acceptance_rate', sampler.avg_acceptance_rate.value
+        print 'final stepsize', sampler.stepsize.get_value()
+        print 'final acceptance_rate', sampler.avg_acceptance_rate.get_value()
 
         return sampler
 
     def test_hmc():
         sampler = sampler_on_nd_gaussian(HMC_sampler.new_from_shared_positions,
                 burnin=1000, n_samples=1000, dim=5)
         assert abs(sampler.avg_acceptance_rate - sampler.target_acceptance_rate) < .1
-        assert sampler.stepsize.value >= sampler.stepsize_min
-        assert sampler.stepsize.value <= sampler.stepsize_max
+        assert sampler.stepsize.get_value() >= sampler.stepsize_min
+        assert sampler.stepsize.get_value() <= sampler.stepsize_max
 
 
 The above code can be run using the command: "nosetests -s code/mcrbm/test\_hmc.py". The output is as follows: