Generating haikus with machine learning
I generated 10 000 random samples in samples.txt. Note that the model was trained on haikus from reddit, so, some will contain foul language. A few highlights:
The United States
is not a state of affairs
in America .
If you have the time
to see what the Bible says ,
then God will exist .
That 's why i 'm not sure
i have the right to express
my own opinion .
And i 'd rather not
be able to work for you
than to be a dick .
It 's a good movie .
My bad , i 'm so sorry you
had to endure it .
It 's a good reason
to be a cool person , but
you can always be .
It 's a pretty good
idea to start a new
job for a while now .
i would n't bother
with any of this if i 'm
not getting a job .
The above represents the top 5%. There are many non-sensical ones for every decent haiku though. Interestingly, the shape is always correct (5-7-5 syllables).
To generate new samples, just run
python generate_haikus.py 3
this prints out 3 haikus in the terminal. See all other options with -h.