This WIP has multiple humble goals:

• Deeper understanding of ML algorithms and structure. Lots of great ML frameworks for sure hidding the painful complexity. I need to understanding the painful complexity to become a better ML dev.
• Level up Math skills.
• Level up Rust programming knowledge.
• Commit into a future where I would use ML and/or Rust as a professional.
• Have fun classifying cat pictures

Steps:

• (IN PROGRESS) Forward and backpropagation framework. API as simple as possible considering the language used. Therefore, lots of macros
• (TO DO) Write convenient input and output nodes.
• (TO DO) Very simple binary image classifier using a single Logistic Regression node in hidden layers.
• (TO DO) Locally storing weights and biases.
• (TO DO) LeNet-5 (implenting other kind of nodes: convolution, pooling, dense etc...).
• (TO DO) Other kind of famous models.
• (TO DO) Letzgongue!

I'm not sure what I'm doing at this point = contains no test 💀

Made simple using macros:

model!(
    input_layer!(convert_img_to_vector("/path/to/file")),
    hidden!(
        layer!(
            10,
            logistic_regression,
        ),
        layer!(
            12,
            logistic_regression,
        )
    ),
    output_layer!(sigmoid)
)
.run()
.train(0.5, y_vector)

Framework automatically compute size of weights and biases with respect to previous layer's size.

Stack of provided (TO DO) activation functions, nodes and classifiers, ready to use. Those can also be specified through closures (TO DO).