# Create the model x = tf.placeholder(tf.float32, [None, NUM_FEATURES]) y_ = tf.placeholder(tf.float32, [None, NUM_CLASSES]) # Build the graph for the deep net hidden_layer = layers.dense(x, 10, activation=tf.nn.relu, kernel_initializer=init.orthogonal( np.sqrt(2)), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-6)) output_layer = layers.dense(hidden_layer, 6, kernel_initializer=init.orthogonal( ), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-6)) cross_entropy = tf.nn.softmax_cross_entropy_with_logits_v2( labels=y_, logits=output_layer) loss = tf.reduce_mean(cross_entropy) + get_regularization_loss() # Create the gradient descent optimizer with the given learning rate. optimizer = tf.train.GradientDescentOptimizer(learning_rate) train_op = optimizer.minimize(loss) correct_prediction = tf.cast( tf.equal(tf.argmax(output_layer, 1), tf.argmax(y_, 1)), tf.float32) accuracy = tf.reduce_mean(correct_prediction) with tf.Session() as sess: sess.run(tf.global_variables_initializer()) train_err = [] test_acc = [] for i in range(epochs):
y_ = tf.placeholder(tf.float32, [None, 1]) # Build the graph for the deep net hidden_layer = layers.dense(x, 30, activation=tf.nn.relu, kernel_initializer=init.orthogonal(np.sqrt(2)), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-3)) output_layer = layers.dense(hidden_layer, 1, kernel_initializer=init.orthogonal(), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-3)) loss = tf.reduce_mean(tf.square(y_ - output_layer)) + get_regularization_loss() #Create the gradient descent optimizer with the given learning rate. optimizer = tf.train.GradientDescentOptimizer(learning_rate) train_op = optimizer.minimize(loss) error = tf.reduce_mean(tf.square(y_ - output_layer)) with tf.Session() as sess: sess.run(tf.global_variables_initializer()) train_err = [] for i in range(epochs): idm = 0 while idm < 1000: nidx = idm + 32 train_op.run(feed_dict={x: trainX[idm:nidx], y_: trainY[idm:nidx]}) idm = nidx
tf.reset_default_graph() with tf.Session() as sess: # Create the model lr = tf.placeholder(tf.float32, []) x = tf.placeholder(tf.float32, [None, NUM_FEATURES]) y_ = tf.placeholder(tf.float32, [None, 1]) # Build the graph for the deep net hidden_layer = layers.dense(x, neuron_count, activation=tf.nn.relu, kernel_initializer=init.orthogonal(np.sqrt(2)), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-3)) output_layer = layers.dense(hidden_layer, 1, kernel_initializer=init.orthogonal(), bias_initializer=init.zeros(), kernel_regularizer=l2_regularizer(1e-3)) loss = tf.reduce_mean(tf.square(y_ - output_layer) ) + get_regularization_loss() # Create the gradient descent optimizer with the given learning rate. optimizer = tf.train.GradientDescentOptimizer(lr) train_op = optimizer.minimize(loss) error = tf.reduce_mean(tf.square(y_ - output_layer)) fold_errors = [] for o in range(NUM_FOLDS): sess.run(tf.global_variables_initializer()) xTrainX = np.split(trainX, NUM_FOLDS, axis=0) xTrainY = np.split(trainY, NUM_FOLDS, axis=0) xValidationX = xTrainX[o] xValidationY = xTrainY[o]