This repo is for playing with markov models/chains, and most likely will evolve into some flavor of twitter bot.
Trainer is used for returning a trained set of data. Currently the list of options for a given key are not calculated as percentages, but return as a list of redundancies.
example:
{'multi token key': ['foo', 'bar', 'bar', 'bar']}
Training sets can be downloaded from google for images at 100 by 100 pixel resolution and cropped to center by running
python create_image_dataset.py '<string>' --count <int> --size <int> <int> --cRange <int>
the string is the work that google will search for images for
- count - number of images to retrieve (currently max 20)
- size - 2 ints for width and height respectively that represent the size of all images in the set
- cRange - range of possible/distributed RGB values to reduce image to
- for example a cRange of 2 would enable a red value of either 0 or 255 where a cRange of 4 would allow 0,85,170,255 as possible values
An image can be generated from any set of images within a directory
by calling
python generate_image.py '<string>' --size <int> <int> --norder <int> --pickle <bool>
- - path to image directory
- size - 2 ints for width and height respectively that represent the size of all images in the set
- norder - what order markov chain to use
- ?pickler - optional takes true/false bool value for whether to load/generate pickle data for set
Stepper Steps through a trained model to create a phrase from its tokens, starting with a random key.