Trainer Engine History of Changes (Release notes)

    Trainer Engine v. 3.1

    New features and improvements:

    • On the automatic model building screen a new slider is added to control the degree of hyperparameter optimisation. With this you can control execution time and the number of models built.
    • Underlying Boruta algorithm implementation was replaced with a more stable one.
    • During descriptor generation, structures can be dropped from the final descriptor set due to various errors. The failure cause is now added to the report.
    • Several performance issues were addressed in this release. Both model building and prediction were improved.

    Trainer Engine v. 3.0

    New features and improvements:

    • Open-source DeepChem machine learning library is available to create training models. Random Forest, Gradient Tree Boost and Graph Convolutional Network algorithms can be used.
    • Predictions can be run by a simple file upload. Descriptors are calculated in the background automatically.
    • Machine Learning problem type can be detected automatically.
    • Automatic DB migration tool from v. 2.0 is available.
    • Old plugin configuration got a proper UI element ('Data processing and confidentiality').
    • Automatic Model Building is extended by DeepChem models.
    • Hyper-parameter names are changes to conventional DeepChem terminology.

    Trainer Engine v. 2.0

    New features and improvements:

    • Open-source descriptors from DeepChem (RDKit and Mordred) are available in Trainer Engine.
    • Descriptor JSON editor improvement: descriptors can be selected from a user-friendly dialogue.
    • Automatic feature selection can be turned off during Automated Model Building.
    • Health and version information box about connected services are added.
    • Descriptors are filtered out based on variance and correlation during automated feature selection.
    • Chemaxon descriptor generation is faster.

    Trainer Engine v. 1.3

    New features and improvements:

    • Filters are introduced on the Runs page.
    • Single runs on the Analyze page can be managed with single-click operations, e.g. add, remove, archive.

    Trainer Engine v. 1.2

    New features and improvements:

    • Comprehensive in-app helps are introduced in the GUI.
    • Advanced options for the automatic descriptor selection ('Use advanced configuration') are introduced.
    • Boruta insights are available on the Details page of a descriptor set.

    Trainer Engine v. 1.1

    New features and improvements:

    • Automated feature selection is introduced.
    • Automated model building is introduced.
    • Training data points and descriptor filtering is introduced in the Analyze page.
    • Observed data distribution for a training model is introduced on the Details page of the model.
    • Runs table paging is introduced.
    • New default descriptor configuration is set.

    Trainer Engine v. 1.0

    New features and improvements:

    • Trainer Engine GUI is available as an option in the CLI.
    • Configuration files in hJSON format are supported by Trainer Engine.
    • k Nearest Neighbour (kNN) descriptor is available for both regression- and classification-type models.
    • Standardization of the training set before descriptor generation is available using Standardizer.
    • The mtryRatio parameter is available in the Random Forest Classification and Regression models.
    • Gradient Tree Boost Classification and Regression models are available.