fcnn

fcnn

fcnn is a scalable fully connected neural network completely written from scratch using the Go Standard Library and an third party library to display the progress bar. fcnn takes advantage of Go's concurrency patterns to parallelize matrix multiplication and all operations a neural network can perform non sequentially. fcnn was also developed following some TDD principles. The GIF shows the binary training a config file with the MNIST dataset, but any other model can be trained as well, by defining a different dataset reader that satisfies the Reader interface.

Usage

Install the binary:

go get github.com/jmpargana/fcnn

You can enhance it with your own dataset readers, they just need to implement two functions DataFrom(images, labels string) ([]Instance, error) and PredictDataFrom(filename string) (matrix.Matrix, error) as described in the readers/reader_type.go file.

License

• MIT License
• Copyright 2020 © João Pargana