// Copyright 2015, University of Freiburg,
// Chair of Algorithms and Data Structures.
// Author: Elmar Haussmann <haussmann@cs.uni-freiburg.de>.

// NOTE: this file should be seen as addition to the TIP file
// of sheet 1. Only the additional methods are shown here.
// The solution of sheet 2 also gives you methods for evaluating
// a benchmark.

// Class for a simple inverted index.
class InvertedIndex

  // This methods first computes the sparse term-document matrix using the
  // (already built) inverted index. For LSI it also performs SVD using
  // dimensionality k and only the m most frequent terms. Intermediate
  // results are stored as members of this class.
  void preprocessing_vsm(int k, int m)

  // Execute the query using the (full) term-document matrix in the vsm.
  // You will need to create a vector from the query and use the term-document
  // matrix from the function above. The result should be exactly the same as
  // when using the inverted index!!! (Easy to verify)
  void process_query_vsm(String query)

  // Execute the query by mapping the query vector to latent space.
  // The term-document matrix in latent space should be precomputed in
  // preprocessing_vsm. Once you have implemented interpolation between
  // original BM25 and LSI lambda controls the combination.
  void process_query_lsi(String query, float lambda)

  // Compute the term-term association matrix T. Write the 50 term pairs with
  // highest values to a file "term_pairs.txt". Only consider unique pairs of
  // different terms.
  void related_term_pairs()

// Main program:
//
// 1. Arguments: <records> <k> <m> <lambda> <benchmark-file>
// 2. Construct inverted index from given file
// 3. Perform VSM preprocessing
// 4. Run benchmark without LSI (only BM25). Print the results.
// 5. Run benchmark only with LSI. Print the results.
// 6. Run benchmark interpolated via lambda with LSI and BM25. Print the results.

void main