guidj / irc Goto Github PK

Recall scale is in the order of 1/100, when it should be 1/10 in the global evaluation plot

The global precision measure is calculated by dividing the sum of the precision for each query over the number of standard values, when it should be divided by the number of queries

After loading a corpus for the first time and creating an index from it, the index object can save (pickled) and loaded next time faster

Evaluate system performance against benchmark from MED.REL

The indeces don't keep the original ID of the document

When comparing/matching documents, the array-index needs to be increment by 1 to match the original document