Beginning Anomaly Detection Using

 Spatial Dropout 1D

keras.layers.SpatialDropout1D()

This function drops entire 1D feature maps instead of neuron elements, but 

otherwise has the same functionality as the regular dropout function. In earlier 

convolutional layers, the feature maps tend to be strongly correlated, so regular dropout 

functions won’t help much with regularization in that case. Spatial dropout helps address 

this and also helps improve independence between the feature maps themselves.

The function takes one parameter:

• 

rate: A float between 0 and 1 that determines the proportion of input 

units to drop.

Figure A-14.  Notice how the flattening layer reduces the dimensionality of its input

Appendix A   intro to KerAs

---

334