enniolearning 0.2.0

A Machine Learning engine for MID music files, based on PyTorch framework.

Ennio-Learning

Ennio is a machine learning program. It takes musics in MID format to learn a theme. Themes are collections of datasets (collection of MID files) having common features, for example :

• Heroic fantasy
• Thriller
• Horror
• Suspense
• Classic
• Manga
• etc.

Once learning (training) for one theme is done, the model (network weights) is serialized and can be reused for later music generation. Such models are used for instance in Ennio app project.

Training features

Ennio Learning handles Jordan features during training and model evaluation. As these are long processes, Jordan app gives remote access to logs (and ETA), and some control over loops. For example, one may skip a training task if it is not going to be efficient (loss is not decreasing).

Logging

A default logger exists, at level logging.INFO. You may adjust this level by calling

import logging
from enniolearning.utils import set_default_logger_level

set_default_logger_level(logging.DEBUG)

You can also provide your own logger (defined from logging library) by passing logger argument. For example :

import logging, logging.config
logging.config.fileConfig('logging.yml')
logger_from_config = logging.getLogger('simpleExample')

from ennio_training import train
train(logger=logger_from_config)