Generates text in the style of Arthur Conan Doyle's Sherlock Canon using deep learning.
"what is the name of the case which had been surely to think that i have ... where does the window of your barrow. it is not a few minutes that it was a station and let me was ... horace harker and the subject of the station which had been surely to think ... the public not unnaturally and in the hall man. i have the dark front of the morning. i should have ... armstrong at the door. "suppose i want to shut hat avoid to chantimage." "and ne! could we have no d ... Run bin/sample.py to generate random samples There are two approaches to sampling :
- Greedy, Large values of temperature make samples more interesting but produces more spelling mistakes
- Beam Search, Values of k=3-4 produce interesting but sometimes grammatically incorrect samples; Values of k=8-10 produce repeatitive but semantically correct samples