start at zero

module 0 builds the autograd engine every network here relies on

read more

machine learning, read from the papers up

every algorithm,
built from scratch

13 modules, 43 papers, numpy only. each page has the paper, the code, and a demo that checks one against the other.







                       00000000000
                  0111111111111111110
              0011111111111111111111111
            01111111111111111111111111110
           111111111110     011111111111110
          11111111110           01111111111
         1111111110               0111111111
        1111111111                 011111111
       01111111110                  11111111
       1111111110                   11111111           0111
     01111111110        01111110  1111111110         011111
 01111111111111          111111111111111110          111111
  1111111111110          01111111111111110          011111
  1111111111111            11111111111110           111111
   1111111111110             111111110             111111
   11111111111110                                01111111
    11111111111110                             011111111
     111111111111110                         0111111111
      11111111111111110                    01111111111
        11111111111111111100            011111111111
         111111111111111111111111111111111111111111
            111111111111111111111111111111111111
              11111111111111111111111111111111
                 11111111111111111111111111
                      1111111111111111
decision regions of the module 11 mlp on two spirals (train acc 1.000); 1 = confident, 0 = near the boundary

the modules