| 19 | from nltk import pos_tag, word_tokenize |
| 20 | |
| 21 | class HMMClassifier: |
| 22 | def __init__(self): |
| 23 | pass |
| 24 | |
| 25 | def fit(self, X, Y, V): |
| 26 | K = len(set(Y)) # number of classes - assume 0..K-1 |
| 27 | N = len(Y) |
| 28 | self.models = [] |
| 29 | self.priors = [] |
| 30 | for k in range(K): |
| 31 | # gather all the training data for this class |
| 32 | thisX = [x for x, y in zip(X, Y) if y == k] |
| 33 | C = len(thisX) |
| 34 | self.priors.append(np.log(C) - np.log(N)) |
| 35 | |
| 36 | hmm = HMM(5) |
| 37 | hmm.fit(thisX, V=V, print_period=1, learning_rate=1e-2, max_iter=80) |
| 38 | self.models.append(hmm) |
| 39 | |
| 40 | def score(self, X, Y): |
| 41 | N = len(Y) |
| 42 | correct = 0 |
| 43 | for x, y in zip(X, Y): |
| 44 | lls = [hmm.log_likelihood(x) + prior for hmm, prior in zip(self.models, self.priors)] |
| 45 | p = np.argmax(lls) |
| 46 | if p == y: |
| 47 | correct += 1 |
| 48 | return float(correct) / N |
| 49 | |
| 50 | |
| 51 | # def remove_punctuation(s): |