{"id":275,"date":"2024-06-28T10:45:02","date_gmt":"2024-06-28T09:45:02","guid":{"rendered":"https:\/\/blogs.imperial.ac.uk\/computing\/?p=275"},"modified":"2024-07-01T11:44:34","modified_gmt":"2024-07-01T10:44:34","slug":"high-speed-rail-data-querying-train-prices-with-c","status":"publish","type":"post","link":"https:\/\/blogs.imperial.ac.uk\/computing\/2024\/06\/28\/high-speed-rail-data-querying-train-prices-with-c\/","title":{"rendered":"High Speed Rail (Data): Querying Train Prices with C++"},"content":{"rendered":"

Hi, I\u2019m Hamish; I\u2019m from Watford, UK, and I\u2019m in the second year of Computing. In my free time I enjoy studying foreign languages, playing water polo and doing Competitive Programming. This year I\u2019m behind Imperial CyberSoc and CTF Team. Let\u2019s connect on\u00a0LinkedIn<\/a>!\"Hamish<\/p>\n

\n

A<\/span> few months <\/span>ago,<\/span> I took <\/span>the <\/span>train to Oxford with a <\/span>school-<\/span>friend. <\/span>Rolling through the <\/span>picturesque <\/span>Chilterns<\/span>, <\/span>the sun <\/span>streaming through the <\/span>window<\/span>,<\/span> and as the fans of <\/span><\/span>GBRJ<\/span><\/span><\/a> that we are, we proposed<\/span> visit<\/span>ing more <\/span>UK cities by train<\/span>. Yet, existing booking sites <\/span>wouldn\u2019t<\/span> let us sort destinations by price.<\/span> Fin<\/span>e<\/span>, we thought: a bit of programming <\/span>will <\/span>solve this<\/span>. <\/span>It also <\/span>happened to <\/span>explain a<\/span> long-running enigma <\/span>– <\/span>why <\/span>UK <\/span>train prices seem so random<\/span>:<\/span><\/span>\u00a0<\/span><\/span><\/p>\n

<\/p>\n

 <\/p>\n

\"Hamish \"\"<\/p>\n

Thanks to Chiltern Railways for (not!) sponsoring this video\u2026<\/span><\/i>\u00a0<\/span><\/p>\n

National Rail<\/span><\/a> publishes open fare data, and various programs have been distributed to process it, from <\/span>librailfare<\/span><\/a> and Avantix Travelller (old data format) to BR Fares (web API unsuitable for our query volume).\u00a0<\/span>\u00a0<\/span><\/p>\n

The most promising implementation I found was @FedericoBotta\u2019s <\/span>railfares<\/span><\/a> which, although using the new data, computes \u2018isocosts\u2019 rather than journey costs. It therefore didn\u2019t seem to consider \u2018split-ticketing\u2019, changing train, or non-reversible fares. I also struggled to make it run quickly on my computer: problematic when examining fares from all tube-accessible stations. Time, then, to write some C++.<\/span>\u00a0<\/span><\/p>\n

Things started off swimmingly with some rudimentary code to import the data, and I ran Dijkstra\u2019s algorithm to compute the cost to every station with unlimited changes. This gave interesting results. I could reach most UK stations for less than \u00a320, even Glasgow for only \u00a314 one-way:<\/span>\u00a0<\/span><\/p>\n

\"\"<\/p>\n

But closer review revealed a snag: my access to virtually all Northern stations hinged on this mysterious \u2018NXP\u2019 ticket to Retford for only \u00a38. The wonderful <\/span>BRFares<\/span><\/a> identified it as LNER\u2019s \u201cflat fare\u201d advance, but alas on their website it was nowhere to be booked. A <\/span>twitter response<\/span><\/a> from customer support confirmed that these are special tickets in limited numbers (often 0!).<\/span>\u00a0<\/span><\/p>\n

This proved a recurring theme as I tried the tickets my program proposed. Almost all cheap rail tickets are quota-controlled \u201cAdvance\u201d fares. While commonly available, they typically have >20 different price points and without the quota (RARS) data, choosing the correct one is impossible.<\/span>\u00a0<\/span><\/p>\n

After much frustration, I resigned to only Anytime and Off-Peak tickets. These are so expensive they are useless, but at least I could check the results:<\/span>\u00a0<\/span><\/p>\n

\"\"<\/p>\n

Furthest stations from London for \u00a3<20, only \u201cwalk up\u201d fares, unlimited changes.<\/span><\/i>\u00a0<\/span><\/p>\n

Another issue with my first attempt was the runtime (>25s). Even if my program didn\u2019t compute quite what I wanted, I could still optimise it to learn from it.<\/span>\u00a0<\/span><\/p>\n

In the end I reduced it to <1s by:\u00a0<\/span>\u00a0<\/span><\/p>\n