def test_d2_1_featvec():
label = '1980s'
fv = clf_base.make_feature_vector({'test': 1, 'case': 2}, label)
eq_(len(fv), 3)
eq_(fv[(label, 'test')], 1)
eq_(fv[(label, 'case')], 2)
eq_(fv[(label, constants.OFFSET)], 1)
def estimate_nb(x, y, smoothing):
"""
Estimate a naive bayes model
:param x: list of dictionaries of base feature counts
:param y: list of labels
:param smoothing: smoothing constant
:returns: weights, as a default dict where the keys are (label, word) tuples and values are smoothed log-probs of P(word|label)
:rtype: defaultdict
"""
labels = set(y)
counts = defaultdict(float)
doc_counts = defaultdict(float)
# Create references to parameters that match formula and algorithm in 4.14 in J&M
# Yes, it violates python naming conventions, but understanding Naive Bayes is more important
D = x # all documents
C = labels # Labels or class as it is known in 4.14 in J&M
V = Counter() # Vocabulary counter
V_size = 0 # |V| = size of vocabulary, initialize to 0
weights = defaultdict()
for c in C:
V = preproc.aggregate_counts(D)
p_xy = estimate_pxy(D, C, c, smoothing, V)
weights.update(clf_base.make_feature_vector(p_xy, c))
# for w, p in p_xy.items():
# key = (c, w)
# weights[key] = p
# Don't really understand how the OFFSET is related to Naive Bayes,
#key = (c, OFFSET)
#weights[key] = smoothing
return weights