To get predictions from trained models you can download model in Deploy view or you can upload test dataset into MLJAR and use it for prediction. This situation is described below.

  1. Upload you data file in Sources view.
  2. Accept attributes usage in Preview for uploaded dataset.
  3. Go to Predict view, in top right corner select dataset which will be used for prediction.
  4. Select algorithm that will be used, by selecting checkbox.
  5. Press Start Prediction and wait a while. Click Refresh on the bottom of the page to see if prediction is computed.
  6. Your predictions will be displayed in the bottom of the page where you can download it.

Predict method

In case of training with cross-validation (CV) all models computed during CV will be used for prediction and their predictions will be averaged. For example, if you run algorithm training with 5 fold CV, then for each parameter setting you have 5 models trained on different train folds. For computing a prediction on new dataset, there will be computed 5 predictions from each model and averaged - this is what you download.

In case of training with separate validation dataset, there is only one model trained and only one model is used for computing predictions.