This man trained an artificial intelligence to generate the most British sounding place names – The indy100

Dan Hon recently set out out to do something rather fun with artificial intelligence.

The director of content at Code for America took a bunch of existing British placenamesand, with a method that goes overour heads somewhat, managed to train an AI to generate some new ones.

His results are so fantasticthat they sound like any number of drizzly English villages you may have driven through looking for a half hidden wedding venue.

Check out Dan's working out, andall the incrediblyconvincing sounding British placenamesbelow.

I trained an A.I. to generate British placenames

The results were predictable.

(Inspired in part by Janelle Shanes New paint colors invented by neural network. Tom Taylor did similar work in 2016, generating English village names.)

Method:

1.Find a list of British placenames. Heres one you can download as a CSV. You just need the names, so strip out all the other columns. To save some time, you can use the one I prepared earlier.

2.Pick a multi-layer recurrent neural network to use. The first time I did this, Karpathys char-rnn was all the rage, this time I used jcjohnsons torch-rnn.

3.If youre using a Mac, dont bother trying to get OpenCL GPU support working. I wasted 3 hours. Just use crisbals CPU-based docker image. (If you know what youre doing, then youre already comfortable doing this all on AWS or youve got an nVidia GPU).

4.Follow jcjohnsons instructions in the readme (pre-process your data, etc.)

5. Go and have a cup of tea while you train your model.

6. Mess around with the temperature when you sample based on your model.

7.Take a look at some of my favourite neural network generated British placenames (and if youd like more, heres 50,000 characters worth):

root@themachine:~/torch-rnn# th sample.lua -checkpoint cv/checkpoint_8450.t7 -length 1000 -gpu -1

Ospley

Stoke Carrston

Elfordbion

Hevermilley

Ell

Elles Chorels

Ellers Green

Heaton on Westom

Hadford Hill

Hambate Combe

Manory Somerstow

Buchraston-on-Ter-Sey

Brotters Common

Normannegg

Twettle Row

North Hill Row St Marne

Torston-le Taney

North Praftton

Tontons Coss

Topswick End

Brumlington

Boll of Binclestead

Farton Green Pear End

Wadworth Mayshyns Wiwton

Wader Bridge

Weston Parpenham

Oarden of Land Park

Batchington Crunnerton

Larebridge Heath Brook

Capton Briins Forehouint Eftte Green

Waryburn Torner Midlwood

Wasts Halkstack

Maggington Common

Stach Helland Neston

Stoke Hills

Sutsy Compton

Stoke of Inch

Upper Somefield

Rastan-on-croan

Wadway Dynd-Rott End

Wattings Ward

Harhester Willey

Marrock

Saxford

Salton Southens Hovers

Salt, Earth

Stamorn Vale

Stouth Wiesleyt Bhampton

Upper Brynton

Kniness Gartes

Webury Hill

Eastbridge Brook

Wallow Manworth

East Holmsley Anby

Hallaid or Humme

Galling Compton

Hampers Hill

Hangyds Hain

Wasland Commone

Wantham Mount on-by Langham

Kinindworthorpe Marmile

Dompton Ole

Dimmer Common

Keston Upper Rhington

Towerhaite Mank

Cockhanford Vales

Porcoft Green

Newtons St Pethen

Silmers Hill

Crocken-ons Clow

Prrighstock Tabergate

Crisklethes Chorn

Cross Gorburster

Storton of Brook

Cartswood Csters

New Amherston

Wascood Woots Corner

West Dottisley

Westovel (Blingwars

Sandeside Backton

Waledon of Bandsead

Rald Bockan-Sea

Boleland Brase

Stoop Heath

See the article here:

This man trained an artificial intelligence to generate the most British sounding place names - The indy100

Related Posts

Comments are closed.