Updated recommenders example

[ThomasDelteil]

Description

Updated recommender system example:

• Merged notebook 1 and 2 into a single one, simpler to follow
• Removed negative sampling notebook, pointing to GluonNLP implementation that contains ImportanceSampling, NoiseContractiveLoss and NegativeSampling.
• Removed manual sparse implementation, replaced with embedding layers
• Updated DSSM model with modernized Gluon version
• Replaced BagOfWordRandomProjection with Gluon version, though it really should only be used in very specific cases, added warning about that in the notebook and used embedding + lstm instead.

[ankkhedia]

@ThomasDelteil Thanks for the contribution
@mxnet-label-bot [ pr-awaiting-testing, Example ]

[ankkhedia]

@aaronmarkham @thomelane @safrooze Could you please take a look ?

[Ishitori]

Cannot comment on notebook - too big, so pasting comments here:
demo1-MF:
1)
ratings = nd.dot(net1.user_embedding.weight.data(ctx=mx.gpu(0)), net1.item_embedding.weight.data(ctx=mx.gpu(0)).T).asnumpy()

Please, replace mx.gpu with ctx[0], so it would work without gpu.

2. net3.summary(user.as_in_context(ctx), item.as_in_context(ctx))
   Line fails with "TypeError: copyto does not support type <class 'list'>", because ctx is an array. Use ctx[0] instead.

3. Same error in Visualizing embeddings - ctx used as a list - needs to be changed to ctx[0]

4. Couldn't plot the last chart (with lines showing losses). Exception was: x and y must have same first dimension, but have shapes (15,) and (2,), and y (loss) is actually of shape (2,)

[ThomasDelteil]

@Ishitori thanks, looks like my last minute changes broke everything :)
Doing one last full run and will push the changes

[ThomasDelteil]

@Ishitori updated

[Ishitori]

Yes, now notebooks work fine. LGTM!

[ThomasDelteil]

@indhub can you please merge this? Thanks!

[kalyc]

@ThomasDelteil thanks for the contribution! Could you re-trigger CI with an empty commit?

[ThomasDelteil]

@indhub @nswamy @sandeep-krishnamurthy can you please merge this? Thanks!

[kalyc]

@mxnet-label-bot update [pr-awaiting-merge]