Here is the original post: https://old.reddit.com/r/rstats/comments/1dz1ve4/how_many_lines_of_code_would_it_take_to_write

I finally found some time to get searching the interwebs for some answers. So here is the situation:

1) Use the flattened MNIST dataset, which is a dataset of 28x28 greyscale images, where values are bounded from 0 to 255 for each pixel. Flattened means instead of a matrix for each image, one row is one image. (so the data is 785 columns = 784 pixels y variable)

2) The only solution that works with flattened data seems to be Keras. I didn't look too deep into Keras, but it does include some unnecessary complexities. It makes sense because it has matured over time.

3) I found a very nice video (only 30m! half explaning math half coding) that explains step by step on the math using only numpy in Python, in matrix format https://www.youtube.com/watch?v=w8yWXqWQYmU

The challenge here is two-fold: 1) convert the matrix math into columnar math and 2) convert the Python code into R (specifically tidy-piped syntax) code.

However, I am having difficulty converting the above video's math matrices into columnar format math. It is a bit complex by hand, but because tidytable includes purrr library functions, I believe it is definitely doable as long as the math is there.

If anyone would like to work with me on converting the math part, I would be grateful and would love to convert it into tidytable code.