Skip to content
プロジェクト

glouw/tinn has 2,138 GitHub stars and 189 forks. GitHot measured -1 stars in 24 hours, -1 in 7 days, and -1 in 15 days. Last checked 2026-11-20 19:20 UTC. Last push 2021-01-17.

この数値の測り方

できること

  • Portable - Runs where a C99 or C++98 compiler is present.
  • Sigmoidal activation.
  • One hidden layer.

計測したスター増加

24時間
-1
7日間
-1
15日間
-1
1か月
履歴を収集中

期間ごとの純増数です。期間は重複しており、日別の値ではありません。未計測の値は空欄です。スター数は注目度の指標であり、品質の保証ではありません。

概要

A tiny neural network library

機械学習

Tinn (Tiny Neural Network) is a 200 line dependency free neural network library written in C99.

For a demo on how to learn hand written digits, get some training data:

wget http://archive.ics.uci.edu/ml/machine-learning-databases/semeion/semeion.data

make; ./test

The training data consists of hand written digits written both slowly and quickly. Each line in the data set corresponds to one handwritten digit. Each digit is 16x16 pixels in size giving 256 inputs to the neural network.

At the end of the line 10 digits signify the hand written digit:

0: 1 0 0 0 0 0 0 0 0 0
1: 0 1 0 0 0 0 0 0 0 0
2: 0 0 1 0 0 0 0 0 0 0
3: 0 0 0 1 0 0 0 0 0 0
4: 0 0 0 0 1 0 0 0 0 0
...
9: 0 0 0 0 0 0 0 0 0 1

This gives 10 outputs to the neural network. The test program will output the accuracy for each digit. Expect above 99% accuracy for the correct digit, and less that 0.1% accuracy for the other digits.

Features

  • Portable - Runs where a C99 or C++98 compiler is present.

  • Sigmoidal activation.

  • One hidden layer.

Tips

  • Tinn will never use more than the C standard library.

  • Tinn is great for embedded systems. Train a model on your powerful desktop and load it onto a microcontroller and use the analog to digital converter to predict real time events.

  • The Tinn source code will always be less than 200 lines. Functions externed in the Tinn header are protected with the xt namespace standing for externed tinn.

  • Tinn can easily be multi-threaded with a bit of ingenuity but the master branch will remain single threaded to aid development for embedded systems.

  • Tinn does not seed the random number generator. Do not forget to do so yourself.

  • Always shuffle your input data. Shuffle again after every training iteration.

  • Get greater training accuracy by annealing your learning rate. For instance, multiply your learning rate by 0.99 every training iteration. This will zero in on a good learning minima.

Disclaimer

Tinn is a practice in minimalism.

Tinn is not a fully featured neural network C library like Kann, or Genann:

https://github.com/attractivechaos/kann

https://github.com/codeplea/genann

Ports

Rust: https://github.com/dvdplm/rustinn

Other

A Tutorial using Tinn NN and CTypes

Tiny Neural Network Library in 200 Lines of Code