python - Probabilities for computers decision in ML? -

phrased quite badly, trying ask how can display percentage of confidence in predicting classification algorithm? using scikit-learn.

say trying identify whether apple or orange based on texture , weight:

#features:  0 = "bumpy" 1 = "smooth" #labels:    0 = apple 1 = orange features = [[140, 1], [130, 1], [150, 0], [170, 0]] labels = [0, 0, 1, 1]  # using decision tree in instance clf = tree.decisiontreeclassifier() clf = clf.fit(features, labels) print(clf.predict([[160, 0]])) 

so predicting [160, 0], based on pattern, , computer predict orange. is there way scikit-learn can predict confidence computer has in returning either 1 or 0? important when have more parameters in feature vector.

yep.

just use predict_proba(x) function (instead of predict()).

probability = clf.predict_proba([[160, 0]]) 

certain classifiers in scikit have ability this, others don't.

in case of decisiontreeclassifier, model, when asked probability of given class, give fraction of elements in training set same class in particular "leaf".

a leaf in decision tree set of conditions (rules) represents path down tree.

for example example [0, 160] represent [x1, x2], rule might have been

if x1 < 10:     if x2 > 150:         # in our training set of `n` examples, 100 fell under          # rule set. 75 of them apple, , 25 orange - thus:         probability = [0.75, 0.25]  # p(apple) = .75, p(orange) = .25 

and of course in binary classification case (two classes) scikit returns both, need 1 or other because probabilities complementary (1 - .75 = .25).

check out docs here learn more.

hope helps.