You want to use the dropout() function in tensorflow.contrib.layers, not the one in tensorflow.nn.

The only differences in the two functions are:

1. Thetf.nn.dropouthas parameterkeep_prob: "Probability that each element is kept"
   tf.layers.dropouthas parameterrate: "The dropout rate"
   Thus,keep_prob = 1 - rate
   as definedhere
2. Thetf.layers.dropouthastrainingparameter: "Whether to return the output in training mode (apply dropout) or in inference mode (return the input untouched)."   The first one turns off(no-op) when not training, which is what you want, while the sec‐ond one does not.