1d Attention Example / recommendation

[ragulpr]

Would be nice with an example to be used with attention. I have experimented but I'm not up to date enough to know what the most popular approach for compression / pruning / regularization is. Preferrably I'd rewrite this section to be useful with attention rather than RNN's as they are out:

taildropout/README.md

Lines 88 to 101 in 5ed14a0

	#### Sequences
	"Recurrent dropout" == Keep mask constant over time. Popular approach.
	```
	x = torch.randn(n_timesteps,n_sequences,n_features)
	gru = nn.GRU(n_features,n_features)
	taildropout = TailDropout(batch_dim = 1, dropout_dim = 2)
	x, _ = gru(x)
	x = taildropout(x)
	```
	If you want to have mask vary for each timestep and sequence
	```
	taildropout = TailDropout(batch_dim = [0,1], dropout_dim = 2)

Ref #10 also see #4