Version 1.0.0
[v1.0.0] - 2024-07-07
Added
-
The
train()
method was added to the 'core.model.DGDModel' class to train the DGD model. -
The
dgd_train
executable was added to train the DGD model using the command line. -
A new type of configuration file containing the options to train the DGD model is available. An example can be found in the newly created
bulkDGD/ioutil/configs/training
directory inside the package. This file, along with the other configuration files, is installed with the package. -
A new example of a configuration file (
model_untrained.yaml
) containing the options to set up the DGD model is available in the 'bulkDGD/ioutil/configs/model' directory for when the model needs to be set up before training. -
The documentation now includes a new tutorial on how to train the DGD model (Tutorial 3).
-
The
load_loss()
andsave_loss()
functions were introduced in the newbulkDGD.ioutil.lossio
module to load and save CSV files containing the losses reported during the training procedure.
API-breaking changes
- The configuration file used to find the representations for new samples now has a new format. Please take a look at the documentation for a detailed explanation of the new format. The format is not backward compatible.
Internal changes (for contributors)
-
The
_get_data_loader()
method has been introduced into thecore.model.DGDModel
class to create data loaders. -
There is now only one internal method in the
core.model.DGDModel
class responsible for optimizing one or multiple representations for a set of samples,_optimize_rep()
. The_get_representations_one_opt()
and the_get_representations_two_opt()
methods have been updated accordingly. -
New sanity checks have been introduced when loading configurations (
ioutil.configio
module). -
A new
_get_final_dataframes_train
method has been introduced to create the data frames produced by the newtrain()
method in thecore.model.DGDModel
class.
Notes
-
The documentation was updated to reflect all changes made to the codebase.
-
Anders Lykkebo-Valløe is now a contributor.
-
Andreas Bjerregaard is now a contributor.