Calculate squad evaluation metrics overall and separately for text answers and no answers

[julian-risch]

Squad evaluation metrics for QA are now calculated a) overall (as before), b) for questions with text answer and c) for questions with no answer.

Questions with no answer are identified by (start,end) == (-1,-1) and the calculation of the metrics is done by splitting the predictions and labels accordingly into two sets.

Also: Fixing a bug that appears when processing ground truth labels where either the first token in the text is the correct (and complete) answer or the very last token. These cases were wrongly handled as impossible_to_answer. Example IDs in dev-v2.json: '57340d124776f419006617bf', '57377ec7c3c5551400e51f09'

Limitations: The number of tokens in a passage passage_len_t and the index of the last token answer_end_t are counterintuitive. There are cases when answer_end_t == passage_len_t.

Closes #686

[julian-risch]

I ran the question_answering_accuracy.py benchmark and can confirm that the numbers are in the same range:
gold_EM = 0.784721
gold_f1 = 0.826671
gold_tnacc = 0.843594 # top 1 recall

'EM': 0.7843005137707403
'f1': 0.8260896852846605
'top_n_accuracy': 0.8430893624189337

[julian-risch]

The benchmark results for no_answer questions are now exactly the same for our implementation and the official squad evaluation. The results for text_answer questions still differ (slightly).

Our evaluation:
'EM': 0.7847216373283922, 'f1': 0.8268405564698051, 'top_n_accuracy': 0.8437631601111766,
'EM_text_answer': 0.7513495276653172, 'f1_text_answer': 0.8357081523221991, 'top_n_accuracy_text_answer': 0.8696018893387314, 'Total_text_answer': 5928,
'EM_no_answer': 0.8179983179142136, 'f1_no_answer': 0.8179983179142136, 'top_n_accuracy_no_answer': 0.8179983179142136, 'Total_no_answer': 5945,

Official squad evaluation:
{
"exact": 79.87029394424324,
"f1": 82.91251169582613,
"total": 11873,
"HasAns_exact": 77.93522267206478,
"HasAns_f1": 84.02838248389763,
"HasAns_total": 5928,
"NoAns_exact": 81.79983179142137,
"NoAns_f1": 81.79983179142137,
"NoAns_total": 5945
}

[Timoeller]

Nice feature, also good catch of the conversion bug. LG!