Entry tags:
Neural networks and fantasy titles, part 2
Go read part 1 before you read this. :)
Anyway, I have discovered the word-level training setting, I'd previously been doing the letter-level setting. I also restricted it to learning from clumps of 5 words together. I think this might produce better titles. Let's see.
Full results: https://docs.google.com/document/d/1e7K5jNWMvz0iadPk6rAgg70nM7WE-S1tYasn92KQwo0/edit?usp=sharing
It all shows up in lowercase with this setting. It's still producing actual titles, some of which I'm sure are in the selections below, but I probably didn't recognize them when copying & pasting.
the dragon ' s kiss
the dragon ' s coda
whisper of fire
dark dark defender ( twelve houses , # 1 )
the darkest passion ( lords of the underworld # 11 )
daughters of winter
daughters of ruin
child of a heartstone
a school : a a a novel
broken man
broken lines
broken empire
blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood
the mermaid ' s sister
the devil ' s apprentice
the pirate ' s daughter
the sorcerer ' s wedding
the romeo catchers
stone and fire ( bastard academy , # 1 )
the battle itself and the string ( the mist of magic for rain wild gate
the throne of the wolfings
the scarlet fig
the curse of palimpsest worlds
the princess that binds
the goblins of the snark
queen of the queen queen of the thief of the rings , part one ( the history of middle - earth , # 7 )
lake of souls ( ( twilight , # 3 )
the face wind in the abyss
the plague diaries
daughter of the first law ( daughter of the fallen , # 8 )
a thousand beginnings and endings
the edge of the crown
brisingr monogatari
the the moon moon moon
to doom of shadows
the darkest passion of the unicorn
the ghost girls and the stone
a wedding and ancient light
the reluctant age of the crowthistle , # 1 )
the serpent bride ( darkglass , # 1 )
the dragons of babel
watership firebird ( malazan book of the fallen , # 3 )
the strange high house in the mist
the lonely beacon
the glass spare
troll story : the magic of cold ( symphony of ages , # 2 )
the phoenix on the sword
the realmgate on the borderland
among wolves
empire wells of the bridge of dawn & the dawn of dawn
I also did another run through the dataset, allowing it to learn from 10 words at a time, but that produced way more gibberish. I'm not going to post it here, but you can see it in this doc:
https://docs.google.com/document/d/1Xif2SIJJQvswfpq1CcV319FsY4egvL9z7vjlrH2kbU0/edit?usp=sharing
Anyway, I have discovered the word-level training setting, I'd previously been doing the letter-level setting. I also restricted it to learning from clumps of 5 words together. I think this might produce better titles. Let's see.
Full results: https://docs.google.com/document/d/1e7K5jNWMvz0iadPk6rAgg70nM7WE-S1tYasn92KQwo0/edit?usp=sharing
It all shows up in lowercase with this setting. It's still producing actual titles, some of which I'm sure are in the selections below, but I probably didn't recognize them when copying & pasting.
the dragon ' s kiss
the dragon ' s coda
whisper of fire
dark dark defender ( twelve houses , # 1 )
the darkest passion ( lords of the underworld # 11 )
daughters of winter
daughters of ruin
child of a heartstone
a school : a a a novel
broken man
broken lines
broken empire
blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood blood
the mermaid ' s sister
the devil ' s apprentice
the pirate ' s daughter
the sorcerer ' s wedding
the romeo catchers
stone and fire ( bastard academy , # 1 )
the battle itself and the string ( the mist of magic for rain wild gate
the throne of the wolfings
the scarlet fig
the curse of palimpsest worlds
the princess that binds
the goblins of the snark
queen of the queen queen of the thief of the rings , part one ( the history of middle - earth , # 7 )
lake of souls ( ( twilight , # 3 )
the face wind in the abyss
the plague diaries
daughter of the first law ( daughter of the fallen , # 8 )
a thousand beginnings and endings
the edge of the crown
brisingr monogatari
the the moon moon moon
to doom of shadows
the darkest passion of the unicorn
the ghost girls and the stone
a wedding and ancient light
the reluctant age of the crowthistle , # 1 )
the serpent bride ( darkglass , # 1 )
the dragons of babel
watership firebird ( malazan book of the fallen , # 3 )
the strange high house in the mist
the lonely beacon
the glass spare
troll story : the magic of cold ( symphony of ages , # 2 )
the phoenix on the sword
the realmgate on the borderland
among wolves
empire wells of the bridge of dawn & the dawn of dawn
I also did another run through the dataset, allowing it to learn from 10 words at a time, but that produced way more gibberish. I'm not going to post it here, but you can see it in this doc:
https://docs.google.com/document/d/1Xif2SIJJQvswfpq1CcV319FsY4egvL9z7vjlrH2kbU0/edit?usp=sharing

no subject
no subject
no subject
no subject
no subject
no subject
no subject
I'm a little worried about all that blood, though. Computer, are you feeling OK?
no subject