Skip to content

Latest commit

 

History

History
118 lines (80 loc) · 3.05 KB

CHANGELOG.md

File metadata and controls

118 lines (80 loc) · 3.05 KB

Imbrium

Released

0.1.0

  • base library

0.1.2

  • encoder-decoder predictor support for multistep multivariate time series

0.1.3

  • added model loss variable type
  • architecture layers can now be adjusted by defining number of neurons and activation function type
  • improved docstrings

0.1.4

  • added keras native call-back feature to fit_model method
  • minor docstring changes
  • minor default value issues in fit_model method solved

0.1.5

  • added getter methods for optimizer and model id
  • added more unit-tests
  • save_model method allows now to specify a customized path

0.1.6

  • solved python version conflict by only allowing python 3.7.*

0.1.7

  • improved source code format
  • improved scaling selector mechanism

0.1.8

  • major refactoring of code base
  • minor name change from Imbrium to imbrium

1.0.0

  • added tests for new utils module
  • name changes of predictor classes:
    • univarstandard => univarpure
    • BasicMultStepUniVar => PureUni
    • univarhybrid => univarhybrid (unchanged)
    • HybridMultStepUniVar => HybridUni
    • multivarstandard => multivarpure
    • BasicMultSTepMultVar => PureMulti
    • multivarhybrid => multivarhybrid (unchanged)
    • HybridMultStepMultVar => HybridMulti
  • tox added
  • outsourced Binder demo notebook to https://github.com/maxmekiska/ImbriumTesting-Demo
  • new README.md

1.0.1

  • imbrium supports now:
    • python 3.7, 3.8, 3.9, 3.10

1.1.0

  • removed batch_size parameter from fit_model method
  • hyperparameter optimization added via the Optuna library

1.2.0

  • added Tensorboard support
  • changed show_performance plot to show loss and metric values
  • added optional dropout and regularization layers to architectures

1.3.0

  • added depth parameter to architectures
  • added optimizer configuration support
  • added optimizer configuration to seeker

2.0.0

  • adapted keras
  • removed internal hyperparameter tuning
  • removed encoder-decoder architectures
  • improved layer configuration via dictionary input
  • split data argument into target and feature numpy arrays

2.0.1

  • fix: removed dead pandas imports
  • chore: added tensorflow as base requirement

2.1.0

  • feat!: removed data preparation out of predictor class, sub_seq, steps_past, steps_future need now to be defined in each model method
    • allows for advanced hyper parameter tuning
  • fix: removed tensor board activation logic bug

3.0.0

  • chore!: changed from temp library keras_core to keras > 3.0.0
  • chore!: removed python 3.8 support to accomodate tensorflow and keras dependiencies
  • chore: increased major to 3.0.0 to align with keras major
  • feat: added evaluate_model method to test model performance on test data
  • refactor!: removed validation split from fit_model. Control validation and test split via evaluation_split and validation_split paramters in class variables

3.1.0

  • feat: added optional batch_size paramter to fit_model
  • feat: added Tensor Board to evaluate_model
  • refactor!: train, test, validation split default change
  • chore: added pre-commit checks
  • refactor: added, improved typing