Fig. 7
From: Predicting enhancers with deep convolutional neural networks
Diagram of data augmentation. Suppose the model accepts sequences of length W bps as input. a In the case that an enhancer is shorter than W, we slide a window of size W along the genome with stride s (default 2) around the input sequence, and take every sequence overlapping with the original one to obtain augmented sequences. b In the case that an enhancer is longer than W, we slide a window of size W along the input sequence with stride s (default 2) to obtain a number of sequences, each of length W