NDCG tutorial notebook

[steaphenai]

Docs: #93

Description

This PR adds the step-by-step Jupyter notebook tutorial for the NDCG metric.

It creates the ndcg-metric-tutorial.ipynb file in the tutorials/intermediate/ directory, as discussed with @vfdev-5. This notebook demonstrates how to effectively use the metric for ranking evaluations and will serve as the destination link for the docstring update in the main pytorch-ignite repository.

Type of Change

• New tutorial / documentation

Checklist

• The Jupyter notebook runs successfully from top to bottom without errors.

[aaishwarymishra]

@steaphenai
I think we should make a function for calculating the dcg score.

For the case for calculating the ideal dcg we can just call the function and send y_true in both parameters, true labels and predicted like
dcg(y_true, y_true, k)

It wll have 2 benefits:

1. No repeated code.
2. User just needs to understand the dcg function once.

currently you are writing 2 loops for doing same thing twice, and the loop style in both case is different .

[steaphenai]

@steaphenai I think we should make a function for calculating the dcg score.

For the case for calculating the ideal dcg we can just call the function and send y_true in both parameters, true labels and predicted like dcg(y_true, y_true, k)

It wll have 2 benefits:

1. No repeated code.
2. User just needs to understand the dcg function once.

currently you are writing 2 loops for doing same thing twice, and the loop style in both case is different .

Yes, I Refactored into a single compute_dcg function and call it for both DCG and IDCG and made the necessary changes. Thanks!

[aaishwarymishra]

looks fine, but I think sorting of scores could be included in dcg method itself, compute_dcg(y_true,y_pred,k) for sorting according to the y_pred, and to get ideal score compute_dcg(y_true,y_true,k), which sorts according to the y_true.

@vfdev-5 thoughts?

Other than that it looks good to me.

[aaishwarymishra]

@vfdev-5 It looks good to me