How to use tensorflow ranking for prediction ?

[vertika-19]

As per my understanding from sample dataset, libsvm generator is assuming that relevance level will be present.

I am using sample
Can you suggest what changes to make so that I can use my trained model later for prediction of relevance levels/ranking based on features and document list for a query.

[eggie5]

This package uses the Tensorflow Estimators API, so accordingly your mode has a predict function. You can use this to rank documents in a query:

for x in ranker.predict(input_fn=lambda: input_fn_eval(test_path)):
    print(x)

[tomleung1996]

This package uses the Tensorflow Estimators API, so accordingly your mode has a predict function. You can use this to rank documents in a query:

for x in ranker.predict(input_fn=lambda: input_fn_eval(test_path)):
    print(x)

I'm new to learning to rank, and I need some help on understanding the model's output.
For each iteration, it outputs a (100,) array for a single query, do you know how to interpret this result?

Does it indicate the relevance score of the first 100 "document" (order by document_id)? If so, how can I find the corresponding document in the test dataset? (it seems we don't specify an document_id in the dataset)

Thanks!

[sbruch]

Vertika - Apologies for the delayed response. Alex is correct; you will need to invoke the learned model in predict mode to obtain per-document score. You may then sort documents by their score to obtain a ranked list.

Tom - generally speaking, the per-query output is a Tensor of relevance scores where the i^{th} score corresponds to the i^{th} document in the input Tensor. Note that during training, the dimension of the input Tensor that corresponds to the number of documents per query is fixed to "list-size" (in your case, it appears to be 100). If there are fewer than list-size documents available for your a query, its Tensor is padded.

Please let us know if you have additional questions.