bug fix

Yusuke Sugomori · Yusuke Sugomori · commit edf9d0a785c4 · 2013-01-05T16:56:21.000+09:00
diff --git a/DeepBeliefNets.py b/DeepBeliefNets.py
@@ -134,7 +134,7 @@ def predict(self, x):
 
 
 def test_dbn(pretrain_lr=0.1, pretraining_epochs=1000, k=1, \
-             finetune_lr=0.1, finetune_epochs=1000):
+             finetune_lr=0.1, finetune_epochs=200):
 
     x = numpy.array([[1,1,1,0,0,0],
                      [1,0,1,0,0,0],
@@ -153,7 +153,7 @@ def test_dbn(pretrain_lr=0.1, pretraining_epochs=1000, k=1, \
     rng = numpy.random.RandomState(123)
 
     # construct DBN
-    dbn = DBN(input=x, label=y, n_ins=6, hidden_layer_sizes=[2], n_outs=2, numpy_rng=rng)
+    dbn = DBN(input=x, label=y, n_ins=6, hidden_layer_sizes=[10], n_outs=2, numpy_rng=rng)
 
     # pre-training (TrainUnsupervisedDBN)
     dbn.pretrain(lr=pretrain_lr, k=1, epochs=pretraining_epochs)
diff --git a/LogisticRegression.py b/LogisticRegression.py
@@ -25,21 +25,21 @@ def __init__(self, input, label, n_in, n_out):
 
         # self.params = [self.W, self.b]
 
-    def train(self, lr=0.1, input=None):
+    def train(self, lr=0.1, input=None, L2_reg=0.00):
         if input is not None:
             self.x = input
 
-        p_y_given_x = softmax(numpy.dot(self.x, self.W) + self.b)
+        p_y_given_x = sigmoid(numpy.dot(self.x, self.W) + self.b)
         d_y = self.y - p_y_given_x
         
-        self.W += lr * numpy.dot(self.x.T, d_y)
+        self.W += lr * numpy.dot(self.x.T, d_y) - lr * L2_reg * self.W
         self.b += lr * numpy.mean(d_y, axis=0)
         
         # cost = self.negative_log_likelihood()
         # return cost
 
     def negative_log_likelihood(self):
-        sigmoid_activation = softmax(numpy.dot(self.x, self.W) + self.b)
+        sigmoid_activation = sigmoid(numpy.dot(self.x, self.W) + self.b)
 
         cross_entropy = - numpy.mean(
             numpy.sum(self.y * numpy.log(sigmoid_activation) +
@@ -50,10 +50,10 @@ def negative_log_likelihood(self):
 
 
     def predict(self, x):
-        return softmax(numpy.dot(x, self.W) + self.b)
+        return sigmoid(numpy.dot(x, self.W) + self.b)
 
 
-def test_lr(learning_rate=0.01, n_epochs=1000):
+def test_lr(learning_rate=0.01, n_epochs=200):
     # training data
     x = numpy.array([[1,1,1,0,0,0],
                      [1,0,1,0,0,0],
@@ -75,7 +75,7 @@ def test_lr(learning_rate=0.01, n_epochs=1000):
     # train
     for epoch in xrange(n_epochs):
         classifier.train(lr=learning_rate)
-        # cost = classifier.negative_log_likelihood(y=y)
+        # cost = classifier.negative_log_likelihood()
         # print >> sys.stderr, 'Training epoch %d, cost is ' % epoch, cost
         learning_rate *= 0.95
 
diff --git a/RestrictedBoltzmannMachine.py b/RestrictedBoltzmannMachine.py
@@ -154,8 +154,8 @@ def test_rbm(learning_rate=0.1, k=1, training_epochs=1000):
 
     for epoch in xrange(training_epochs):
         rbm.contrastive_divergence(lr=learning_rate, k=k)
-        cost = rbm.get_reconstruction_cross_entropy()
-        print >> sys.stderr, 'Training epoch %d, cost is ' % epoch, cost
+        # cost = rbm.get_reconstruction_cross_entropy()
+        # print >> sys.stderr, 'Training epoch %d, cost is ' % epoch, cost
 
 
     # test
