https://github.com/kentonishi/ml-number-detection
A slow and unoptimized number detection neural network.
https://github.com/kentonishi/ml-number-detection
Last synced: 3 months ago
JSON representation
A slow and unoptimized number detection neural network.
- Host: GitHub
- URL: https://github.com/kentonishi/ml-number-detection
- Owner: KentoNishi
- Created: 2019-03-30T03:09:45.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2019-04-30T17:20:22.000Z (about 6 years ago)
- Last Synced: 2025-01-03T10:11:42.939Z (5 months ago)
- Language: C++
- Homepage:
- Size: 34.7 MB
- Stars: 3
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Machine Learning Number Detection
A *slow and unoptimized* number detection neural network.
## About
This project uses code from [Simple Neural Network](https://github.com/KentoNishi/Simple-Neural-Network/tree/v1) and [Bitmap Numbers Dataset](https://github.com/KentoNishi/Bitmap-Numbers-Dataset).
## Details
All images in this repository are resized to ``16x16`` with 24 bit color.
Images of each digit are converted into the ``data.in`` format for the neural network. The pixel values are encoded into a list of floating point numbers in the following fashion:
```
1-(((r+g+b)/3)/255)
```
This creates a value for each pixel that is always between 0 and 1.
The list is generated by scanning the image left to right, top to bottom.
The last number in the list indicates the expected result.
## Files
The ``data`` subdirectory contains the samples used for creating the dataset, as well as extra C++ scripts. These undocumented scripts are used for resizing images and converting files.
## Issues
* Configuration parameters were manually selected, so they may be highly inefficient.
* The code uses stochastic gradient descent, causing speed issues.
* The code uses sigmoid activation functions, causing speed issues.
* The tests are identical to the dataset, causing almost 100% accuracy.
* Large datasets do not run in sufficient time.