python - Layerwise learning rate implementation in Tensorflow -

i'm trying implement answer 2 given following question on stackoverflow: how set layer-wise learning rate in tensorflow? seek use specific learning rate first 2 layers , rate 10 times less third , final layer.

these weights:

 weights = {     # 9x9 conv, 1 input, 64 output feature maps     'wc1': tf.variable(tf.random_normal([9, 9, 1, 64])),     # 1x1 conv, 64 inputs, 32 output feature maps     'wc2': tf.variable(tf.random_normal([1, 1, 64, 32])),     # 5x5 conv, 32 inputs, 32 output feature maps     'wc3': tf.variable(tf.random_normal([5, 5, 32, 1]))     #'wc3': tf.variable(tf.random_normal([33*33*32, 33*33]))     }  biases = {         'bc1': tf.variable(tf.random_normal([64])),         'bc2': tf.variable(tf.random_normal([32])),         'bc3': tf.variable(tf.random_normal([1]))     } 

and implementation follows:

 self.opt = tf.train.gradientdescentoptimizer(learning_rate=learning_rate)  self.grads_and_vars = self.opt.compute_gradients(self.cost, [ self.weights['wc1'], self.biases['bc1'], self.weights['wc2'], self.biases['bc2'], self.weights['wc3'], self.biases['bc3'] ])    # apply gradients  train_op = self.opt.apply_gradients([self.grads_and_vars[0], self.grads_and_vars[1], self.grads_and_vars[2], self.grads_and_vars[3],          (self.grads_and_vars[4]/10, self.weights['wc3']), (self.grads_and_vars[5]/10, self.biases['bc3'])], global_step=self.global_step) 

unfortunately, following error when running code pertains train_op operator (apply_gradients):

  valueerror: shapes must equal rank, 4 , 5 'gradientdescent/update_variable_2/applygradientdescent' (op: 'applygradientdescent') input shapes: [5,5,32,1], [], [20,5,5,32,1].  

i cannot understand tensor shape [20,5,5,32,1] since expecting [5,5,32,1] tensor weights. missing?