So far all the examples I can find for Tensorflow Ranking only take sparse features, I already have a data set pre-processed to numpy array in following format:
relevance, query, feature_1, feature_2, ... feature_100
2, "a", 0.1, 0.2, ... 0.15
1, "a", 0.2, 0.21, ... 0.15
I'm looking at this example, where it reads in libsvm data and returned a feature dictionary where each key is the feature column name: https://github.com/tensorflow/ranking/blob/master/tensorflow_ranking/examples/tf_ranking_libsvm.py#L136
But then it gets flattened to a dimension of [query_size * list_size, feature_count] before feeding in the graph.
https://github.com/tensorflow/ranking/blob/master/tensorflow_ranking/examples/tf_ranking_libsvm.py#L323
Do I have to reformat my data in order to use tensorflow ranking?
I see example_feature_columns is used in encode_listwise_features and encode_pointwise_features, Is creating example_feature_columns a must for tf ranking?