In the lecture on `Metrics for evaluating performance`

, the video describes the mean absolute percentage error (MAPE) as the ratio of the prediction error to the predictors, but I think this should be the ratio of the prediction error to the true value being predicted.

In other words:

```
errors = forecasts - actual
mape = np.abs(errors / x_valid).mean()
```

should be:

```
errors = forecasts - actual
mape = np.abs(errors / actual).mean()
```

The keras docs for `MeanAbsolutePercentageError`

seem to agree.