Open christopher5106 opened 6 years ago
Hi,
I would like to implement a layer dropout (not simple Dropout but full LayerDropout). For that, I need a random number generator operator and a switch operator, I believe. Here is the implementation in tensorflow:
pred = tf.random_uniform([]) < self.dropout x_train = tf.cond(pred, lambda: residual, lambda: tf.nn.dropout(x, 1.0 - self.dropout) + residual)
How can I achieve the same in CNTK ?
Thank you for your answer
According to the, still opened, issue #1371 we can not random sampling from a distribution in CNTK.
Hi,
I would like to implement a layer dropout (not simple Dropout but full LayerDropout). For that, I need a random number generator operator and a switch operator, I believe. Here is the implementation in tensorflow:
How can I achieve the same in CNTK ?
Thank you for your answer