A repository for fast, easy, and readable neural network development. Inspired by netharn
- Class prototype that implements typical NN model functionality:
- hyperparameters:
- input shape
- output len
- nice name [ ] initializer (TODO)
- Work directory
- reset
- verbosity
- hyperparameters:
- Provides cache saving and loading built in through the ubelt caching module
Backbone of the entire package, designed to serve as a simple and intuitive superclass for network implementations. Includes the following functionality (subject to change):
parameters(self)
to(self, device)
on_epoch(self, epoch, error)
cache(self)
load_cache(self)
to_onehot
compute_error
deprecated
isiterable
check_constraints
Container class for registering and managing constraints, defaults, and current parameter values.
- Checks values against defined constraints
- sets uninitialized parameters to default values
- produces hashable string for caching
- supports unsetting parameters
shape_for_shape
output_shape_for