Change Log

0.4.0 beta => 0.5.0 beta

  • In the model setting screen, it became possible to select scaling for the data set.
  • When scaling is set, the prediction result, loss value, and evaluation value have been changed to specifications that are displayed as rescaled (returned value) values.

0.3.0 beta => 0.4.0 beta

  • It became possible to training and prediction with data.pickle (pred.pickle) including character strings (half-width alphabetic characters). (Performs training (prediction) by automatically extracting only numeric columns. The prediction result CSV contains strings columns)
  • Added Model Filter function that can narrow down displayed models by Dataset type.
  • In the Model Map section, changed the point of the model being trained to be easy to understand.
  • Added the Best Epoch Line function.
  • Added the display switching function of training data and validation data in the Prediction Sample section.
  • On the Prediction page, it is now possible to compare graphs with predicted results and training data.
  • Added graph display function of Feature Importance.

0.2.0 beta => 0.3.0 beta

  • After completion of model creation, the importance of each explanatory feature can be checked.
  • Random Forest and XGBoost were added to the algorithm selection item at model creation.

0.1.0 beta => 0.2.0 beta

  • Explanatory features and target features can be selected when creating a dataset.
  • Feature Scaling can be selected when creating a dataset.
  • The prediction result can be saved as CSV data.

0.0.0 beta => 0.1.0 beta

  • GraphCNN is possible with GPU.
  • Enabled install with wheel.
  • Enabled starting server from current directory.
  • Enabled training user defined model from GUI.
  • Added regressor module that can pull model to python script from GUI server.