update

Classification/train.py

@@ -0,0 +1,102 @@
+
+
+
+
+
+import numpy as np
+
+
+
+
+
+#Load data
+print 'Loading data'
+with open(home+'/Documents/MNIST_data/mnist.pkl','rb') as f:
+    mnist_data = pickle.load(f)
+
+train_x = mnist_data[0][0]
+train_y = mnist_data[0][1]
+valid_x = mnist_data[1][0]
+valid_y = mnist_data[1][1]
+test_x = mnist_data[2][0]
+test_y = mnist_data[2][1]
+
+
+print (train_x.shape)
+print (train_y.shape)
+print (valid_x.shape)
+print (valid_y.shape)
+
+
+fsadfsa
+
+
+#Load model
+
+
+
+
+
+
+
+#Train model
+random_seed=1
+rs=npr.RandomState(random_seed)
+n_datapoints = len(train_y)
+arr = np.arange(n_datapoints)
+
+if path_to_load_variables == '':
+    self.sess.run(self.init_vars)
+
+else:
+    #Load variables
+    self.saver.restore(self.sess, path_to_load_variables)
+    print 'loaded variables ' + path_to_load_variables
+
+#start = time.time()
+for epoch in range(epochs):
+
+    #shuffle the data
+    rs.shuffle(arr)
+    train_x = train_x[arr]
+    train_y = train_y[arr]
+
+    data_index = 0
+    for step in range(n_datapoints/batch_size):
+
+        #Make batch
+        batch = []
+        batch_y = []
+        while len(batch) != batch_size:
+            batch.append(train_x[data_index]) 
+            one_hot=np.zeros(10)
+            one_hot[train_y[data_index]]=1.
+            batch_y.append(one_hot)
+            data_index +=1
+
+        # Fit training using batch data
+        _ = self.sess.run((self.optimizer), feed_dict={self.x: batch, self.y: batch_y, 
+                                                self.batch_size: batch_size})
+
+        # Display logs per epoch step
+        if step % display_step == 0:
+
+            cost,pred = self.sess.run((self.cost,self.prediction), feed_dict={self.x: batch, self.y: batch_y, 
+                                                self.batch_size: batch_size})
+            # cost = -cost #because I want to see the NLL
+            print "Epoch", str(epoch+1)+'/'+str(epochs), 'Step:%04d' % (step+1) +'/'+ str(n_datapoints/batch_size), "elbo=", "{:.6f}".format(float(cost))#,logpy,logpW,logqW #, 'time', time.time() - start
+            # print 'target'
+            print ["{:.2f}".format(float(x)) for x in batch_y[0]] 
+            # print 'prediciton'
+            # print pred.shape
+            print ["{:.2f}".format(float(x)) for x in pred[0]] 
+            print
+
+if path_to_save_variables != '':
+    self.saver.save(self.sess, path_to_save_variables)
+    print 'Saved variables to ' + path_to_save_variables
+
+
+
+
+

RL4/rl_mar2018_6_implicitoptions_withbuffer/README.md

@@ -0,0 +1,8 @@
+
+Plotting the value function error
+
+
+
+
+
+