Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
Sign up to join this community
from sklearn.naive_bayes import GaussianNB,MultinomialNB
xx = [[1],[1],[1],[2],[2],[3]]
yy = [1,1,1,0,0,0]
clf = GaussianNB()
# clf = MultinomialNB()
clf.fit(xx,yy)
clf.predict(xx)
The expected result is [1,1,1,0,0,0] but code output is [0,0,0,0,0,0].
There is only one feature for xx.
For a MultinomialNB, the modelled probability of the feature per class $p(x_i|y)$ is always the same ($\frac{3}{3}=1$ for class $1$ and $\frac{7}{7}=1$ for class $0$). The prediction is therefore the class prior $p(y_i)$ which is also the same here ($p(0) = 0.5$ and $p(1)=0.5$) so by default class $0$. Having one feature does not really makes sense because it gives no information to the model on what gives discriminative power toward the classes.
For a GaussianNB this is different because the modelled probability is given by $P(x_i \mid y) = \frac{1}{\sqrt{2\pi\sigma^2_y}} \exp\left(-\frac{(x_i - \mu_y)^2}{2\sigma^2_y}\right)$ so the mean and variance per class of the single feature give sufficient information.
