scitkit-learn query data dimension must match training data dimension

Question

I'm trying to use this code from the scikit learn site:

http://scikit-learn.org/stable/auto_examples/classification/plot_classifier_comparison.html

I'm using my own data. My problem is, I have a lot more than two features. If I want to "expand" the features from 2 to 3 or 4....

I'm getting:

"query data dimension must match training data dimension"

def machine():
with open("test.txt",'r') as csvr:

    reader= csv.reader(csvr,delimiter='\t')

    for i,row in enumerate(reader):

        if i==0:
            pass
        elif '' in row[2:]:
            pass
        else:
            liste.append(map(float,row[2:]))

a = np.array(liste)
h = .02 
names = ["Nearest Neighbors", "Linear SVM", "RBF SVM", "Decision Tree",
         "Random Forest", "AdaBoost", "Naive Bayes", "LDA", "QDA"]
classifiers = [
    KNeighborsClassifier(1),
    SVC(kernel="linear", C=0.025),
    SVC(gamma=2, C=1),
    DecisionTreeClassifier(max_depth=5),
    RandomForestClassifier(max_depth=5, n_estimators=10, max_features=1),
    AdaBoostClassifier(),
    GaussianNB(),
    LDA(),
    QDA()]



X = a[:,:3]
y = np.ravel(a[:,13])

linearly_separable = (X, y)
datasets =[linearly_separable]
figure = plt.figure(figsize=(27, 9))
i = 1

for ds in datasets:
    X, y = ds

    X = StandardScaler().fit_transform(X)
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.4)

    x_min, x_max = X[:, 0].min() - .5, X[:, 0].max() + .5
    y_min, y_max = X[:, 1].min() - .5, X[:, 1].max() + .5
    xx, yy = np.meshgrid(np.arange(x_min, x_max, h),
                         np.arange(y_min, y_max, h))

    cm = plt.cm.RdBu
    cm_bright = ListedColormap(['#FF0000', '#0000FF'])
    ax = plt.subplot(len(datasets), len(classifiers) + 1, i)

    ax.scatter(X_train[:, 0], X_train[:, 1], c=y_train, cmap=cm_bright)

    ax.scatter(X_test[:, 0], X_test[:, 1], c=y_test, cmap=cm_bright, alpha=0.6)
    ax.set_xlim(xx.min(), xx.max())
    ax.set_ylim(yy.min(), yy.max())
    ax.set_xticks(())
    ax.set_yticks(())
    i += 1

    for name, clf in zip(names, classifiers):
        ax = plt.subplot(len(datasets), len(classifiers) + 1, i)
        print clf.fit(X_train, y_train)
        score = clf.score(X_test, y_test)
        print y.shape, X.shape
        if hasattr(clf, "decision_function"):
            Z = clf.decision_function(np.c_[xx.ravel(), yy.ravel()])
            print Z
        else:
            Z = clf.predict_proba(np.c_[xx.ravel(), yy.ravel()])[:, 1]


        Z = Z.reshape(xx.shape)

        ax.contourf(xx, yy, Z, cmap=cm, alpha=.8)
        ax.scatter(X_train[:, 0], X_train[:, 1], c=y_train, cmap=cm_bright)

        ax.scatter(X_test[:, 0], X_test[:, 1], c=y_test, cmap=cm_bright,
                   alpha=0.6)

        ax.set_xlim(xx.min(), xx.max())
        ax.set_ylim(yy.min(), yy.max())
        ax.set_xticks(())
        ax.set_yticks(())
        ax.set_title(name)
        ax.text(xx.max() - .3, yy.min() + .3, ('%.2f' % score).lstrip('0'),
                size=15, horizontalalignment='right')
        i += 1

figure.subplots_adjust(left=.02, right=.98)
plt.show()

In this case I use three features. What am I doing wrong in the code, Is it something with the X_train and X_test data? With just two features, everything is ok.

my X value:

(array([[ 1.,  1.,  0.],
   [ 1.,  0.,  0.],
   [ 1.,  0.,  0.],
   [ 1.,  0.,  0.],
   [ 1.,  1.,  0.],
   [ 1.,  0.,  0.],
   [ 1.,  0.,  0.],
   [ 3.,  3.,  0.],
   [ 1.,  1.,  0.],
   [ 1.,  1.,  0.],
   [ 0.,  0.,  0.],
   [ 0.,  0.,  0.],
   [ 0.,  0.,  0.],
   [ 0.,  0.,  0.],
   [ 0.,  0.,  0.],
   [ 0.,  0.,  0.],
   [ 4.,  4.,  2.],
   [ 0.,  0.,  0.],
   [ 6.,  3.,  0.],
   [ 5.,  3.,  2.],
   [ 2.,  2.,  0.],
   [ 4.,  4.,  2.],
   [ 2.,  1.,  0.],
   [ 2.,  2.,  0.]]), array([ 1.,  1.,  1.,  1.,  0.,  1.,  1.,  0.,  1.,  1.,  0.,  1.,  1.,
    1.,  1.,  1.,  0.,  1.,  1.,  0.,  1.,  0.,  1.,  1.]))

The first array is the X array and the second array is the y(target) array.

I'm sorry for the bad format = error:

        Traceback (most recent call last):

File "allM.py", line 144, in <module>
mainplot(namePlot,1,2)
File "allM.py", line 117, in mainplot

Z = clf.predict_proba(np.c_[xx.ravel(), yy.ravel()])[:, 1]

File "/usr/local/lib/python2.7/dist-packages/sklearn/neighbors/classification.py", line 191, in predict_proba
neigh_dist, neigh_ind = self.kneighbors(X)

File "/usr/local/lib/python2.7/dist-packages/sklearn/neighbors/base.py", line 332, in kneighbors
return_distance=return_distance)

File "binary_tree.pxi", line 1298, in sklearn.neighbors.kd_tree.BinaryTree.query (sklearn/neighbors/kd_tree.c:10433)

ValueError: query data dimension must match training data dimension

and this is the X array without putting him into the Dataset "ds".

[[ 1.  1.  0.][ 1.  0.  0.][ 1.  0.  0.][ 1.  0.  0.][ 1.  1.  0.][ 1.  0.  0.][ 1.  0.  0.][ 3.  3.  0.][ 1.  1.  0.][ 1.  1.  0.][ 0.  0.  0.][ 0.  0.  0.][ 0.  0.  0.][ 0.  0.  0.][ 0.  0.  0.][ 0.  0.  0.][ 4.  4.  2.][ 0.  0.  0.][ 6.  3.  0.][ 5.  3.  2.][ 2.  2.  0.][ 4.  4.  2.][ 2.  1.  0.][ 2.  2.  0.]]

Answer 1

This is happening because clf.predict_proba() requires an array where each row has the same number of elements as the rows in the training data -- in other words an input with shape (num_rows, 3).

When you were working with two-dimensional exemplars this worked because the result of np.c_[xx.ravel(), yy.ravel()] is an array with two-element rows:

print np.c_[xx.ravel(), yy.ravel()].shape
(45738, 2)

These exemplars have two elements because they're created by np.meshgrid which the sample code uses to create a set of inputs to cover a two-dimensional space which will plot nicely. Try passing an array with three-item rows to clf.predict_proba and things should work fine.

If you want to reproduce this specific piece of sample code, you'll have to create a 3D meshgrid, as described in this question on SO. You'll also have plot the results in 3D, where mplot3d will serve as a good starting point, though based on the (admittedly brief) look I gave to the plotting in the sample code, I suspect this may be more trouble than it's worth. I'm not really sure how a 3D analog of those plots even look.