Implement K Neighbors Classifier in scikit-learn with 3 feature per object

Question

I would like to implement a KNeighborsClassifier with scikit-learn module (http://scikit-learn.org/dev/modules/generated/sklearn.neighbors.KNeighborsClassifier.html)

I retrieve from my image solidity, elongation and Humoments features. How can i prepare these datas for training and validation? I must create a list with the 3 features [Hm, e, s] for every object i retrieved from my images (from 1 image have more objects)?

I read this example(http://scikit-learn.org/dev/modules/generated/sklearn.neighbors.KNeighborsClassifier.html):

X = [[0], [1], [2], [3]]
y = [0, 0, 1, 1]
from sklearn.neighbors import KNeighborsClassifier
neigh = KNeighborsClassifier(n_neighbors=3)
neigh.fit(X, y) 

print(neigh.predict([[1.1]]))
print(neigh.predict_proba([[0.9]]))

X and y are 2 features?

samples = [[0., 0., 0.], [0., .5, 0.], [1., 1., .5]]
from sklearn.neighbors import NearestNeighbors
neigh = NearestNeighbors(n_neighbors=1)
neigh.fit(samples) 

print(neigh.kneighbors([1., 1., 1.])) 

Why in first example use X and y and now sample?

Solution

Your first segment of code defines a classifier on 1d data.

X represents the feature vectors.

[0] is the feature vector of the first data example
[1] is the feature vector of the second data example
....
[[0],[1],[2],[3]] is a list of all data examples, 
  each example has only 1 feature.

y represents the labels.

Below graph shows the idea:

• Green nodes are data with label 0
• Red nodes are data with label 1
• Grey nodes are data with unknown labels.

    print(neigh.predict([[1.1]]))

This is asking the classifier to predict a label for x=1.1.

    print(neigh.predict_proba([[0.9]]))

This is asking the classifier to give membership probability estimate for each label.

Since both grey nodes located closer to the green, below outputs make sense.

    [0] # green label
    [[ 0.66666667  0.33333333]]  # green label has greater probability

The second segment of code actually has good instructions on scikit-learn:

In the following example, we construct a NeighborsClassifier class from an array representing our data set and ask who’s the closest point to [1,1,1]

>>> samples = [[0., 0., 0.], [0., .5, 0.], [1., 1., .5]]
>>> from sklearn.neighbors import NearestNeighbors
>>> neigh = NearestNeighbors(n_neighbors=1)
>>> neigh.fit(samples) 
NearestNeighbors(algorithm='auto', leaf_size=30, ...)
>>> print(neigh.kneighbors([1., 1., 1.])) 
(array([[ 0.5]]), array([[2]]...))

There is no target value here because this is only a NearestNeighbors class, it's not a classifier, hence no labels are needed.

For your own problem:

Since you need a classifier, you should resort to KNeighborsClassifier if you want to use KNN approach. You might want to construct your feature vector X and label y as below:

X = [ [h1, e1, s1], 
      [h2, e2, s2],
      ...
    ]
y = [label1, label2, ..., ]

Comments

[postgres]

Last question: How can i choose labels? Why it is [0,0,1,1] and not [0,1,2,3] or [0,0,0,1] ?

[postgres]

I got errors: "setting an array element with a sequence" because fit function wants: X : {array-like, sparse matrix, BallTree, cKDTree}. I got a list of list! If i insert in feature vectors all the 7 humoments: "query data dimension must match BallTree data dimension" i tried also to covert it in numpy array, same error : "setting an array element with a sequence."

[greeness]

The label is [0,0,1,1] not [0,1,2,3] because in that example we only have 2 classes not 4 classes. It could be [0,0,0,1] if the three left-most nodes are actually the same class. Can you post your code in a different post for your second question?

[postgres]

but i have 292 object, how can i choose number of classes? well i post my code in a new question

[greeness]

what is your goal of the classification? Are those images describe different topic/pattern/color/object? If yes, for each type, you give a different label. either 0,1,2 or 'red','white','black', the values of the label do not matter. what matters is how many values they are. i also answered the other post you made.