Spam filter - step 2

Create function compute_confusion_matrix() that will compute and return a confusion matrix based on real classes of emails, and on email classes predicted by a filter.

• In the article Binary Classification, find and understand the meaning of abbreviations TP, FP, TN, FN.
• Take a piece of paper and write down:
  • what these abbreviations mean for the spam filtering problem, and
  • what we need to know to be able to compute them.
• See the documentation for namedtuple.

Task:

• In module quality.py, create function compute_confusion_matrix().
• The function will have 4 input arguments:
  • truth_dict, a dictionary with the true correct class of individual emails,
  • pred_dict, a dictionary with the class predicted for individual emails by a filter,
  • pos_tag (optional, with default value True), a class that will be considered positive, and
  • neg_tag (optional, with defualt value False), a class that will be considered negative. Thanks to these optional parameters, the function will be generally usable, not only for the spam filter task with pos_tag=“SPAM” and neg_tag=“OK”).
• The function will compute four statistics, TP, TN, FP, FN, needed to evaluate a filter, and will return them as a namedtuple with the following definition:

  from collections import namedtuple
   
  ConfMat = namedtuple('ConfMat', 'tp tn fp fn')

Why do we need it?

• Function compute_confusion_matrix() represents the basis for evaluation of the filter performance.

The function can be used in the following way. First, an example where both the input dictionaries are empty, i.e. we have no information about any email.

>>> cm1 = compute_confusion_matrix({}, {})
>>> print(cm1)
ConfMat(tp=0, tn=0, fp=0, fn=0)

In the following code, each of TP, TN, FP, FN cases happens exactly once:

>>> truth_dict = {'em1': 'SPAM', 'em2': 'SPAM', 'em3': 'OK', 'em4':'OK'}
>>> pred_dict = {'em1': 'SPAM', 'em2': 'OK', 'em3': 'OK', 'em4':'SPAM'}
>>> cm2 = compute_confusion_matrix(truth_dict, pred_dict, pos_tag='SPAM', neg_tag='OK')
>>> print(cm2)
ConfMat(tp=1, tn=1, fp=1, fn=1)

And in the last example, the predictions perfectly match the real classes, such that only TP and TN are nonzero:

>>> truth_dict = {'em1': 'SPAM', 'em2': 'SPAM', 'em3': 'OK', 'em4':'OK'}
>>> pred_dict = {'em1': 'SPAM', 'em2': 'SPAM', 'em3': 'OK', 'em4':'OK'}
>>> cm2 = compute_confusion_matrix(truth_dict, pred_dict, pos_tag='SPAM', neg_tag='OK')
>>> print(cm2)
ConfMat(tp=2, tn=2, fp=0, fn=0)

Of course, the input dictionaries may have a different number of items than 4.