char-rnn
Character level language model using RNN.
Reference: https://gist.github.com/karpathy/d4dee566867f8291f086 https://github.com/karpathy/char-rnn
Data sources:
- Leo Tolstoy's War and Peace (3.3 MB): ref
- Sherlock Homes full text (3.4 MB): ref 1 / ref 2
- Shakespeare's texts (4.6 MB): ref
- Linux kernel (6.2 MB): ref
- Wikipedia text (100 MB): ref
- Entire Linux source code (474 MB): ref 1 / ref 2
- Arxiv abstracts (46 kB): ref
$ git clone https://github.com/torvalds/linux.git
$ cd linux
$ find . -name "*.[c|h]" | shuf | xargs cat > linux.txt
min-char-rnn output
iter 2500, loss: 64.751443
-----
cam's we ofowkopris, coulst ,
A auss hong t'an af youkz, bes,
Rty, ic-
H Cout, :of woucd ungr pgivow: cave your sircer hics, I secese, Iuc uitf son court soo natt hure sor,
Br hory med Wiun:
CIRUTUS:
-----
iter 2600, loss: 64.002116
-----
fhpooj shs, kyifs thom it uotmurd'd se care.
BNUS:
Toumhe t andius; t, yoor, jon? bp yethes bly youias aur, i, io, d
wns;t'gorn thavanr. wocigivepnoo rresptin hiuvteemor he'envone hinsler rowvorame
-----
iter 2700, loss: 63.210558
-----
el hank the weheld wowthad gltto that mo hathich ve phe pomy, wupe wore the sle the thintdatis nsm horl:
Have yhe yory ul p er,
th't to thit; yor or. ftS
Brimiot pames pes hos entaohet petire me hher
-----
iter 7206300, loss: 44.407445
-----
Vork'ty tword,
And a so the inseinge leve
Compolt and Cay, cosit wither would iset to that nies there honour of is wourger, myet.
If think your me!
Sidge--
Wisefray I a mest came tay untise it,
A me
-----
iter 7206400, loss: 44.328364
-----
thinking and they smears are dints my lymer' or for the shile'd well that I Ho'en you,
The poyas's me it Rest wized to dnabyour, will but'a?
Neghnee larget.
DUKE ANDIK:
Dors.
GLOUCESTER:
What!
You
-----
iter 7206500, loss: 44.499381
-----
Thosowarth'
Acolly you quanss.
TRAILE:
To for him loved oftoued?
Rungtices I slight arree;
I beace her you well, arce have desent, for in sees papest your teave
sholl thistefays, sir: priemace Arros
-----