Alternative weight normalisation

[barneydobson]

In graph theory, it's common practice to set edge weights to values higher than 1. It depends on the application, but in this case, when weight doesn't carry specific physical meaning, and it's just a cost measure in path traversal, why normalize between 0-1 and not 1-5, or something like that. It doesn't even have to be a float, it can be an integer that acts as a classifier. So, as @dalonsoa said, it's better to prevent this numerical issue from occurring instead of ad-hoc solutions.

Right, my main point is that normalizing weight is not as straight forward as just using a min-max normalizer. Since the weight is not a physical property, you'd have to be careful about how does the normalizer interpret different values. In my methodology, I opted for a discrete normalizer, so I can clearly assign a physical meaning to weight. For example, for slope, we can use low, mild, and steep as descriptors for physical interpretation, then based on a pre-defined range bin the values and assign weights of 1, 2, and 3. This way, when the slope difference is in the order 1e-3, values might fall into the same category.

Originally posted by @cheginit in #218 (comment)

Normalise between 1 and 5 instead of 0 and 1 - or better yet, let a user specify the normalisation range. Can also add option to normalise with standard deviation and mean