why and where we should use tf.expand_dims()?

[BhawitBalodi]

why and where we should use tf.expand_dims()?
Please can anyone tell the logic behind this.

[ishandandekar]

Hey @BhawitBalodi , when training the model, the code forms batches of data to make training computationally faster. This means each training batch gets converted into small mini-batches. Each of these mini-batches have shape as (mini_batch_size, ...). This mini-batch size is often a power of 2 (sometimes 32 or 64). So, if you then want to check the prediction for a single example, you first have to match the number of dimensions with these mini-batched examples. This is why we use tf.expand_dims(). It essentially adds a dimension to your tensor. So, if you have a tensor of shape (10, ), and then pass this tensor in the tf.expand_dims() function, you will get a tensor of shape (10, 1). Know more from this link