# Multi-Class Classification from Single-Class Data with Confidences

## Requirements
- Python 3.6
- numpy 1.14
- PyTorch 1.1
- torchvision 0.2

## Demo
The following demo will show the results of SC-Conf Learning with the MNIST dataset. When running the code, the test accuracy of each epoch will be printed for `SC/Sub-Conf`, `Weighted`, and `NoRSC-Conf` in order. The results will have two columns: epoch number and test accuracy.
 

Before running `demo.py`, we can choose the type of confidence. If we run the following code:

```bash
python demo.py -noise True 
```
the confidences are generated in the way mentioned in Remark 1 and thus the `SC/Sub-Conf` is skipped. The default value of `-noise` is False.

The classes used for training can be manually determined by changing the list `ordered_class` in `demo.py`. Any subset of the collection of number 0-9 is acceptable. The default value is [7, 9].