Scatter plot of handwritten digits

Question

I want to create a scatter plot of handwritten digits of 0 and 1 (http://yann.lecun.com/exdb/mnist/). I took 4 samples ie two 0's and two 1's.

Each handwritten digits are having pixel values having dimension 1 cross 778 matrix.

Now I want to do scatterplot of these digits in two dimension to check if they are separable or not.

I tried plotting those in R but the final plot is just mixture of points. How shall I plot in order that each digit's scatterplot looks different from other. am I proceeding in right direction while plotting?

Please advise.

Answer 1

1. They are probably separable, so you don't really need to check it.

2. If you want to know whether they are separable, the right way is to use a classifier such as SVM.

   You are trying to reduce the dimensions of the 4 digits (4x778) into two dimensions (4x2). There are many ways to do it and none is perfect. After you reduce the dimensions some information is always lost, so it is just an approximate way of telling whether the high-dimensional data is separable.

   The most common/simple way to reduce the data to two (or any number of) dimensions is called PCA.