[Tip] Remember epsilon when you make RMSE using Pytorch

## RMSE problem
- RMSE is supposed to be like this.
```python
rmse = torch.sqrt(torch.nn.MSELoss(logits, y.float()))
```
 - __But__, `torch.sqrt(0)` return `NaN`

## Solution
- So, you should add any extremely small number aka `epsilon`
Here's the solution

```python
eps = 1e-6
rmse = torch.sqrt(torch.nn.MSELoss(logits, y.float()) + eps)
```