move sparsity related discussion to another section.

[Ankush-Chander]

Many thanks for this comprehensive guide on embeddings.

My understanding behind proposed change:
Sparsity related discussion at the beginning of section 3.2.2 TF-IDF gives an impression that we solve sparsity issue by moving from one-hot encoding to TF-IDF. However TF-IDF encoding doesn"t change the number of zero cells we have in one-hot encoding, it only effects the weightage given to the non-zero cells, thereby leaving the sparsity of the one-hot matrix unchanged.

Proposed change:

1. Elicit advantage of TF-IDF over one-hot encoding in section 3.2.2 TF-IDF.
2. Move sparsity discussion to section 3.2.3 where matrix factorisation is discussed.

[micaleel]

@Ankush-Chander, your definition of sparsity based on the size of embeddings is debatable . When comparing two embeddings of equal size, the one with fewer empty or zero values is considered dense.

[Ankush-Chander]

Hi @micaleel
Sorry for not being more explicit before. My interpretation of sparsity is exactly as you just mentioned.
Number of cells in term-document matrix that have value 0.

[veekaybee]

Left one small change for clarification, once that's done this is good to go. Thanks for opening this!

[veekaybee]

LGTM!