finish session

src/model.py

@@ -65,9 +65,6 @@ def next(self):
     ox = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))
     om = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))
     ob = tf.Variable(tf.zeros([num_nodes]))
-    # Classifier weights and biases
-    w = tf.Variable(tf.random_uniform([num_nodes, num_nodes]))
-    b = tf.Variable(tf.zeros([num_nodes, 1]))
 
     def lstm_cell(i, o, state):
         input_gate = tf.sigmoid(tf.matmul(ix, i) + tf.matmul(im, o) + ib)
@@ -81,17 +78,17 @@ def lstm_cell(i, o, state):
     output = tf.zeros([num_nodes, 1])
     state = tf.zeros([num_nodes, 1])
     i = 0
+    MSE = 0
     while i < (days-1):
         output, state = lstm_cell(tf.reshape(x[:, i], [5, 1]), output, state)
         y.append(output)
+        MSE += tf.reduce_mean(tf.square(output - tf.reshape(x[:, i+1], [5, 1])))
         i += 1
     x = x[:, 1:]
 
-    logits = tf.matmul(w, tf.reshape(tf.reduce_sum(tf.concat(0, y), 0), [5, 1])) + b
-    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, tf.reshape(tf.reduce_sum(tf.concat(0, x), 1), [5, 1])))
+    loss = MSE/(days-1)
     optimizer = tf.train.AdamOptimizer().minimize(loss)
     yT, _ = lstm_cell(tf.reshape(x[:, i], [5, 1]), output, state)
-    prediction = tf.matmul(w, yT) + b
 
 num_steps = 2000
 summary_frequency = 100
@@ -101,11 +98,17 @@ def lstm_cell(i, o, state):
     mean_loss = 0
     for step in range(num_steps):
         batch = train_batches.next()
+        batch = batch.astype(np.float32)
+        print(batch.shape)
         print(batch.dtype)
         _, l = sess.run([optimizer, loss], feed_dict={x: batch})
         mean_loss += l
         if (step+1) % summary_frequency == 0:
             mean_loss = mean_loss/summary_frequency
+            print('Average MSE at step %f: %s' % (step, mean_loss))
             valid_batch = valid_batches.next()
+            valid_batch = batch.astype(np.float32)
+            v_l = sess.run([loss], feed_dict={x: valid_batch})
+            print('Validation MSE: %f' % v_l)
 
     save_path = tf.train.Saver().save(sess, '../models/' + stock + '_model.ckpt')