Someone has used a neural network to generate new London Tube Stations and you’d believe they were real
Rob Manuel – occasional Poke writer and general bot maker has been fiddling with neural networks and has used one to create new tube stations.
Tube stations generated via neural network. Tweet: hand-picked. Image: unedited.
Brongers Crost
Tonth Street
Hallerston
Harnon Green
Chathal Hark
Hall Hill
Parkington
Surth Park
Barkons Cross
Street Hark
Crosterlad
Nouth Greens
Curth Wicker
Congor Street
North Carling pic.twitter.com/oqgjVItPlI— Rob Manuel (@robmanuel) March 13, 2018
And people have been enjoying playing along
I got out at Brongery Street once. Never again.
— Phil Edwards (@Flandyke) March 14, 2018
It’s mostly hipsters in Hallerston, accountants in Chamsmerth, poshos in St Jant’s Cross and don’t go out at night in Bringe.
— Peter Franklin (@peterfranklin_) March 13, 2018
Bringe gets a bad rep but its got some great bars just south of Sthemen's Green
— Gary Noble (@chenobble) March 14, 2018
We got in touch to ask him more: can you generate some more just for us?
Sure. I’ll just run the code again. How many do you want?
Fansill Oak
Cinge Raak
Canton Road
Blern Crost
Clapham Park
Mannar Green
Crostes Park
Consrett
Kansing
Linessill Stuthfark
Marnnar Breen
Tombrepton
Townenslay
Herthal Crous
Crossharbury
Cancham Crors
Cusham Green
Norton Larken
Sortherboad
Brikget
Park Greens
Surmen Street
Famingal
Stichfordes
Kals’s Green
Northwidge
Stotford Rivars
Halneslam Greens
Sh.Jatess Acth
Sotbury Bark
Touth Bark
Ersthal Broakes
Carkensill
Bronton Road
Greens Algert
Blacknan
Greens Street
Bustona Hill
Bandwor Crren
Crosn Park
Cronserey
Hallesbury
Cringe
Laide Oad
Harnhwary
Shaphers Eak
Fanton Layton
Tottesbury Croas
Stemenss Road
Crennist Linel
Rodpor
Brers Crors
Cresters
Hannhrart
St.aaceswam
Roadens Wcatt
Clasham Green
North Streen
Cepding Lanersal
Barkasd Ruar
Caness Carte
Radens Croste
Lephincurs
Kancron
Central Crrest
Lonersim Street
Ans Toonth
Surtwording
Stuen Park
Tonth Park
Erst Park
Kinghal
Alders Histon
Seuth Harners
Suth Share
North Mlork
North Greens
Can you generate absolutely anything?
As long as you’ve got a lump of source data you can give it a go. Here’s an attempt I made at creating new ZX Spectrum games.
ZX Spectrum games generated by neural net.
Speed Tart
The Neat Wars
Super Frong
Gurons of the Enger [128K]
Spide Race
Fright Baster
The Pudger
Amester 3D
Tomeard Spider
Jungle Racking – The Storm
The Death of Goregor [128K] pic.twitter.com/5DfSmVjT0s— Rob Manuel (@robmanuel) March 13, 2018
If we wanted to do this ourselves – how would be do it?
OK. Do you want a quick tutorial. Right here’s a brain dump.
1. I’m running this in the cloud via a EC2 Amazon instance. They have free tier ones set up with most of the packages you need called things like “Deep Learning AMI (Ubuntu) Version 5.0 – ami-7336d50e”
2. The code itself is on github and called tensorflow-char-rnn
HAVE FUN POKE READERS! MAKE SURE YOU TELL US IF YOU HAVE A PLAY WITH ANYTHING LIKE THIS YOURSELVES.