Found an edge case where flawed generation of RDF triplets can break the evaluation script

[zachares]

In the file Evaluation_script_json.py in the function evaluaterefcand() on line 224 the inputs are supposed to rdf triples i.e. when you run the lines below the function header newreference = reference.split(' | ') and newcandidate = candidate.split(' | '), newcandidate and newreference should be lists of strings of length 3. If either of them are not, then this can cause downstream functions to fail when instead they should give performance values of 0 for that example; so I propose changing the logic on line 229 to

    if len(newreference) != 3:
        newreference = ['', '', '']
    if len(newcandidate) != 3:
        newcandidate = ['', '', '']

And changing the logic of the for loop below to for idx in range(3):. I would have liked to introduce a PR with these suggested changes so they are easier to review, but I cannot.