guess the correlation data inspection a pretty normal distribution train/val/test split splitting amount .dataset: 150000 instances ├─8