PyTorch Char-RNN

PyTorch implementation of a character-level recurrent neural network. See accompanying blog post.

Includes pretrained models for generating:

• fake book titles in different genres
• first names in different languages
• constellation names in English and Latin

Examples

Book titles

Possible categories in the pretrained model include: Adult_Fiction, Erotica, Mystery, Romance, Autobiography, Fantasy, New_Adult, Science_Fiction, Biography, Fiction, Nonfiction, Sequential_Art, Childrens, Historical, Novels, Short_Stories, Christian_Fiction, History, Paranormal, Thriller, Classics, Horror, Philosophy, Young_Adult, Contemporary, Humor, Poetry, Dark, Lgbt, Religion

Heart in the Dark (Romance)

Book of the Dark (Fantasy) 

Bed Store (Young Adult)

Growing Me (New Adult)
Me the Bean (New Adult) 

King of the Dark (Erotica)
Your Mind (Erotica)

Red Story (Mystery) 

Be the Life (Biography)

First names

Possible categories in the pretrained model include: Arabic, Chinese, Czech, Dutch, English, French, German, Greek, Irish, Italian, Japanese, Korean, Polish, Portuguese, Russian, Scottish, Spanish, Vietnamese

Russian:

Rovakov
Uanten
Shantovov

Chinese:

Chan
Hang
Iun

Constellations

English:

orogane
quale
rowans
serpent
kelescop

Latin:

bearis
corac
serpens
xer
zeriscase

File Overview

• generate_books.py: Train/generate fake book titles.
• generate_names.py: Train/generate fake first names.
• generate_constellations.py: Train/generate fake constellation names.
• csv_to_txt.ipynb: Notebook to convert
• models/: Saved models.
• data/: Training data sets as text files.

Setup

All you really need is PyTorch.

Usage

At the bottom of each script, the function call samples(CATEGORY, START_LETTERS) can be modified for whatever category/starting letters you want.

$ python generate_names.py:

• samples('Russian', 'RUS)

$ python generate_books.py:

• samples('Fiction', 'abcdefghijklmnopqrstuvwxyz')

$ python generate_constellations.py

• samples('latin', 'abcdefghijklmnopqrstuvwxyz')

Training your own

Set line 174 and 175 to:

rnn = RNN(n_letters, 128, n_letters)
# rnn = torch.load('models/book_titles.pt')

Uncomment out this part in the script:

print('Training model...')
for iter in range(1, n_iters + 1):
    output, loss = train(*random_training_example())
    total_loss += loss

    if iter % print_every == 0:
        print('%s (%d %d%%) %.4f' % (timeSince(start), iter, iter / n_iters * 100, loss))

### Save model after training
torch.save(rnn, 'language_names.pt')

You can also set the training data directory on line 31: data_glob = 'data/names/*.txt' to data_glob = 'data/your_dir/*.txt'

and setting each category of text within its own .txt file within the data/your_dir/ directory.

Generating your own

If you just want to generate samples, you can comment out the training snippet and set the network to load the pretrained model. i.e. Set line 174 and 175 to:

# rnn = RNN(n_letters, 128, n_letters)
rnn = torch.load('models/book_titles.pt')

and re-run the script.

Credits

• PyTorch tutorial
• Goodreads data scrape