I am trying out to create an Ranker using the REST API after successfully Adding documents to the collection. Do I need to use train.py. If so then whats the use of create Ranker API. Also while trying to create RANKER, can you please tell me where do I need to specify cluster id and collection name. Do I need to specify it in metadata.json file.
Any help would be highly appreciated.
See here for an overview of the methods available for training a ranker: https://www.ibm.com/watson/developercloud/doc/retrieve-rank/training_data.html#methods. Using the REST API directly is described under the 'Manually training a ranker' section and is intended for advanced users that might be appending additional columns to the feature vectors that are generated automatically when you use train.py.
The create Ranker REST API call does not take 'cluster id' and 'collection names' as input parameters. It assumes that you've already used the /fcselect REST API call (which does take 'cluster id' and 'collection id') to pre-generate feature vectors similar to what train.py is doing internally. Again, in the advanced use case, you might take these feature vectors and then augment them using some other custom features as described in this blog post: https://medium.com/machine-learning-with-ibm-watson/developing-with-ibm-watson-retrieve-and-rank-part-3-custom-features-826fe88a5c63#.unfm2ocik
