Hacker News new | past | comments | ask | show | jobs | submit login

transit (the company) has some really nice blog posts about the difficulty of making transit maps



I really like the approach described in these links. Targeted algorithms solve specific problems (e.g. ILP to decide ordering of adjacent lines) instead of someone trying to put together a Grand Unified Machine Learning Algorithm that ends up having to be fudged manually after it gets 99 cases right and makes an egregious mistake in the 100th one.

The more general problem Transit seems to be trying to solve (that no one else is) is having a map that can be kept up-to-date with transit alerts and diversions with minimal hassle. For example, it looks like their launch-time picture of Toronto's streetcars (in the first link) was generated when there was trackwork at the College and Bathurst intersection and the corresponding streetcar lines were being diverted around it.

The second link explains rather well the reasons behind some of the problems evident in Google's Toronto map. For example, the unsnapping issue for adjacent lines occurs a lot because streetcar lines are heavily interlined.

they really are a great company. when there was an outage in san francisco's schedule reporting system, they had users tag themselves as being on specific buses, and tried to synthesise their own real-time schedule from that.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact