Regarding Attention Size

[andyjung12]

The paper said that the attention size is 64, but on the config file, the attention size is 32. Which one is correct?

[pquochuy]

The value in the config is just the default value. The actual value is passed via the command in the bash script file. If you look at, for example, https://github.com/pquochuy/xsleepnet/blob/master/sleepedf-20/tensorflow_nets/xsleepnet1/run_1chan.sh the value 64 is passed to the parameter seq_attention_size1.