#OCR
OCR app made primarily in octave
Client made in python
(Client looks horrible so I am not uploading it yet)
- Speed: It is too slow, Any suggestions would be appreciated
- Comments: I tried commenting each block, but most of it still doesn't make sense to people who are reading it for the first time
- Documentation: No documentation (yet)
- Works only for single digits, should be solved by March
- Works only for BW images, should be solved by January
- Works for very small images, should be solved by January
- Need to upload Python Client, should be solved whenever I make NN1 public, because client is client for both, and crashes if has access to only one...
- Does not have a sanity check for over-fitting or under-fitting.
- Accuracy: 95.66% last time I checked, which was 8/12/2017 8:55 AM
Sorry for the tardiness in uploading all this code because I was just super lazy and never used github... I still need to upload a lot of other projects...
- Overfitting/Underfitting: this'll be solved whenever NN1 goes public because of my overfitting/underfitting checks in those programs (basically checks for smoothness, % accuracy, general shape of the curve, etc.)
- Bad Suggestion for accuracy: "Use more internal levels or Use more nodes per level", I would like to but would like to prevent overfitting
- Upload client: It'll happen when I make NN1 public