Can i train model from Output layer ?

[yusufani]

Hi , I am currently doing some experiments to increase the accuracy of the model. As you know, we have a 1024 lenght array before making a class prediction. I got that array , made some manipulation and created 1024 lenght arrays . I want to continue the training process on the pre-trained model with 1024 lenght arrays but I couldn't find how I could do it. Can you help me ?

    conv1_filter = tf.Variable(tf.truncated_normal(shape=[3, 3, 3, 64], mean=0, stddev=0.08))
    conv2_filter = tf.Variable(tf.truncated_normal(shape=[3, 3, 64, 128], mean=0, stddev=0.08))
    conv3_filter = tf.Variable(tf.truncated_normal(shape=[5, 5, 128, 256], mean=0, stddev=0.08))
    conv4_filter = tf.Variable(tf.truncated_normal(shape=[5, 5, 256, 512], mean=0, stddev=0.08))

    # 1, 2
    conv1 = tf.nn.conv2d(x, conv1_filter, strides=[1,1,1,1], padding='SAME')
    conv1 = tf.nn.relu(conv1)
    conv1_pool = tf.nn.max_pool(conv1, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
    conv1_bn = tf.layers.batch_normalization(conv1_pool)

    # 3, 4
    conv2 = tf.nn.conv2d(conv1_bn, conv2_filter, strides=[1,1,1,1], padding='SAME')
    conv2 = tf.nn.relu(conv2)
    conv2_pool = tf.nn.max_pool(conv2, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')    
    conv2_bn = tf.layers.batch_normalization(conv2_pool)
  
    # 5, 6
    conv3 = tf.nn.conv2d(conv2_bn, conv3_filter, strides=[1,1,1,1], padding='SAME')
    conv3 = tf.nn.relu(conv3)
    conv3_pool = tf.nn.max_pool(conv3, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')  
    conv3_bn = tf.layers.batch_normalization(conv3_pool)
    
    # 7, 8
    conv4 = tf.nn.conv2d(conv3_bn, conv4_filter, strides=[1,1,1,1], padding='SAME')
    conv4 = tf.nn.relu(conv4)
    conv4_pool = tf.nn.max_pool(conv4, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
    conv4_bn = tf.layers.batch_normalization(conv4_pool)
    
    # 9
    flat = tf.contrib.layers.flatten(conv4_bn)
    flat =   tf.identity(flat, name='flat')

    # 10
    full1 = tf.contrib.layers.fully_connected(inputs=flat, num_outputs=128, activation_fn=tf.nn.relu)
    full1 = tf.nn.dropout(full1, keep_prob)
    full1 = tf.layers.batch_normalization(full1)
    full1 = tf.identity(full1, name='full1')
    
    # 11
    full2 = tf.contrib.layers.fully_connected(inputs=full1, num_outputs=256, activation_fn=tf.nn.relu)
    full2 = tf.nn.dropout(full2, keep_prob)
    full2 = tf.layers.batch_normalization(full2)
    full2 = tf.identity(full2, name='full2')
    
    # 12
    full3 = tf.contrib.layers.fully_connected(inputs=full2, num_outputs=512, activation_fn=tf.nn.relu )
    full3 = tf.nn.dropout(full3, keep_prob)
    full3 = tf.layers.batch_normalization(full3)
    full3 = tf.identity(full3, name='full3')    
    
    # 13
    full4 = tf.contrib.layers.fully_connected(inputs=full3, num_outputs=1024, activation_fn=tf.nn.relu)
    full4 = tf.nn.dropout(full4, keep_prob)
    full4 = tf.layers.batch_normalization(full4)  
    full4 = tf.identity(full4, name='full4')     
    
    # 14
    out = tf.contrib.layers.fully_connected(inputs=full4, num_outputs=10, activation_fn=None)
    out = tf.identity(out,name="out")
    return out

You can see the names of the layers in the code.