Deep Learning with PyTorch

Batches

Processing the entire data as a sequence is quite challenging and not computationally efficient. So, we break the sequence data into multiple batches, and treat each as a separate sequence. Though it may not sound straightforward, it works a lot better, as the model can learn quicker from batches of data. Let's take the example where the English alphabet is sequenced and we split it into batches.

Sequence: a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z.

When we convert the preceding alphabet sequence into four batches, we get:

a g m s y

b h n t z

c i o u

d j p v

e k q w

f l r x

In most of the cases, we would end up trimming the last extra words or tokens that form a small batch, since it doesn't have ...