| 15 | from scipy.stats import multivariate_normal as mvn |
| 16 | |
| 17 | class NaiveBayes(object): |
| 18 | def fit(self, X, Y, smoothing=1e-2): |
| 19 | self.gaussians = dict() |
| 20 | self.priors = dict() |
| 21 | labels = set(Y) |
| 22 | for c in labels: |
| 23 | current_x = X[Y == c] |
| 24 | self.gaussians[c] = { |
| 25 | 'mean': current_x.mean(axis=0), |
| 26 | 'var': current_x.var(axis=0) + smoothing, |
| 27 | } |
| 28 | self.priors[c] = float(len(Y[Y == c])) / len(Y) |
| 29 | |
| 30 | def score(self, X, Y): |
| 31 | P = self.predict(X) |
| 32 | return np.mean(P == Y) |
| 33 | |
| 34 | def predict(self, X): |
| 35 | N, D = X.shape |
| 36 | K = len(self.gaussians) |
| 37 | P = np.zeros((N, K)) |
| 38 | for c, g in iteritems(self.gaussians): |
| 39 | mean, var = g['mean'], g['var'] |
| 40 | P[:,c] = mvn.logpdf(X, mean=mean, cov=var) + np.log(self.priors[c]) |
| 41 | return np.argmax(P, axis=1) |
| 42 | |
| 43 | |
| 44 | if __name__ == '__main__': |