Use machine learning on epigenomic data.
To install in dev/editable mode:
- Clone the git
- Create a virtual environment, activate it
- In the root directory (setup.py), run "
pip install -e ." - Install other requirements with pip (see requirements folder).
See input-format folder for examples of mandatory files.
usage: main.py [-h] [--offline] [--predict] [--model MODEL]
category hyperparameters hdf5 chromsize metadata logdir
positional arguments:
category The metatada category to analyse.
hyperparameters A json file containing model hyperparameters.
hdf5 A file with hdf5 filenames. Use absolute path!
chromsize A file with chrom sizes.
metadata A metadata JSON file.
logdir Directory for the output logs.
optional arguments:
-h, --help show this help message and exit
--offline Will log data offline instead of online. Currently cannot merge comet-ml offline
outputs.
--predict Enter prediction mode. Will use all data for the test set. Overwrites hparameter file
setting. Default mode is training mode.
--model MODEL Directory from which to load the desired model. Default is logdir.
The Metadata class has an api to support modification of metadata as needed, like the select_category_subsets and remove_category_subsets methods. One can use them if they want to adjust label values on the fly, or perform additional filtering.
As soon as a label category exists in a dataset, any value is considered the label. Be it "", "--" or "NA".
If datasets containing differents keys is expected or possible, be sure to run remove_missing_labels on the relevant categories.
For additional information, refer to the documentation