nn-rs

A traditional 3 layers neural network with Rust

MNIST handwritten digit database

original format: http://yann.lecun.com/exdb/mnist/

converted csv format: https://pjreddie.com/projects/mnist-in-csv/

How to use it to recognize the handwritten digit?

./target/release/handwritten-digit-recognition ./dataset/mnist_train <path-of-image>

NOTE: the prefix name "2828_my_own" images are from https://github.com/makeyourownneuralnetwork/makeyourownneuralnetwork/tree/master/my_own_images the prefix name "handwrite" images are from mine created from Windows Paint

The Performance of prefix name "2828_my_own" images is better than the prefix "handwrite" images. I think that is because the digit in the the prefix name "2828_my_own" images are more bold than the digit in the prefix name "handwrite" images.

MNIST in CSV

The format is:

label, pix-11, pix-12, pix-13, ... , pix-nn \n
newlabel, pix-11, pix-12, pix-13, ... , pix-nn \n
...

where pix-ij is the pixel in the i-th row and j-th column.

pix-nn is 28*28 = 784 pixel values in the range 0-255 gray values.

label is the digit represented by the image.

For the curious, this is the script to generate the csv files from the original data

def convert(imgf, labelf, outf, n):
    f = open(imgf, "rb")
    o = open(outf, "w")
    l = open(labelf, "rb")

    f.read(16)
    l.read(8)
    images = []

    for i in range(n):
        image = [ord(l.read(1))]
        for j in range(28*28):
            image.append(ord(f.read(1)))
        images.append(image)

    for image in images:
        o.write(",".join(str(pix) for pix in image)+"\n")
    f.close()
    o.close()
    l.close()

convert("train-images-idx3-ubyte", "train-labels-idx1-ubyte",
        "mnist_train.csv", 60000)
convert("t10k-images-idx3-ubyte", "t10k-labels-idx1-ubyte",
        "mnist_test.csv", 10000)

FAQ

Q: Why need to seperate the train data set and test data set?

A: That is we want to test before we train the model. Otherwise, we can let network to remember the training data set and get a high accuracy. But it is not a good model. So that it is normal case in machine learning to seperate the train data set and test data set.

References

• https://github.com/makeyourownneuralnetwork/makeyourownneuralnetwork
• https://makeyourownneuralnetwork.blogspot.com/
• 《Make your own neural network》 by Tariq Rashid