I am trying to implement SVM Classifier over MNIST dataset. As my parameters are 3 dimensional its throwing the following error:
ValueError: Found array with dim 3. Expected <= 2
Following is my code snippet:
import mnist
from sklearn import svm
training_images, training_labels = mnist.load_mnist("training", digits = [1,2,3,4])
classifier = svm.SVC()
classifier.fit(training_images, training_labels)
Does sklearn support a multi-dimensional classifier?
One option for fixing the problem would be to reshape the input data into a 2-dimensional array.
Let's assume that your training data consists of 10 images which are each represented as an 3x3 matrix and therefore your input data is 3-dimensional.
[ [[1,2,3], [[1,2,3], [
[4,5,6], [4,5,6], image 10
[7,8,9]] , [7,8,9]] , ... , ] ]
We can turn each image into an array of 9 elements in order to convert the dataset into 2-dimensions.
dataset_size = len(training_images)
TwoDim_dataset = dataset.reshape(dataset_size,-1)
This would turn the data into the following shape:
[ [1,2,3,4,5,6,7,8,9] , [1,2,3,4,5,6,7,8,9] , ... , [image 10] ]
The problem is with your input data.
You can use sklearn to load a digit dataset as well:
from sklearn.datasets import load_digits
from sklearn import svm
digits = load_digits()
X = digits.data
y = digits.target
classifier = svm.SVC()
classifier.fit(X[:1000], y[:1000])
predictions = classifier.predict(X[1000:])
