Table 1
Characteristics of similar packages. The symbol ‘?’ means that the feature is not officially mentioned in the documentation but it may work. However, the results may not be accurate.
| PACKAGES | ||||||||
|---|---|---|---|---|---|---|---|---|
| multiviewstacking [4] (Python) | mvlearn [8] (Python) | multiview [9] (Python) | multiview [10] (R) | mvlearn R [11] (R) | multi-view- AE [12] (Python) | scikit-multimodallearn [13] (Python) | ||
| Characteristics | Allows choosing any underlying model | ✓ | ✓ | |||||
| Allows passing underlying custom models | ✓ | ? | ? | |||||
| Allows heterogeneous underlying models | ✓ | ✓ | ||||||
| Supports supervised learning | ✓ | ✓ | ✓ | ✓ | ||||
| Supports semi-supervised learning | ✓ | |||||||
| Supports unsupervised learning | ✓ | ✓ | ✓ | ✓ | ||||
| Implements multi-view stacking | ✓ | |||||||
Figure 1
Overall process of Multi-View Stacking at prediction time. The predicted classes of each view are fed to the meta-learner, which produces the final prediction.
Figure 2
Overall steps to build a multi-view stacking model. 1) Define the first-level learners and meta-learner. 2) Instantiate a MultiViewStacking object. 3) Fit the model. 4) Make predictions on new data.
Figure 3
Example documentation text. In a Jupyter Notebook (Windows), the documentation of a function can be displayed by pressing SHIFT+TAB.
Figure 4
First rows and columns of the HTAD dataset.
Listing 1
Load the HTAD dataset.
Listing 2
Define the learners.
Listing 3
Instantiate the multi-view stacking model.
Listing 4
Train the model.
Listing 5
Make predictions on the test set.
Listing 6
Evaluate view 1 independently.
Listing 7
Evaluate view 2 independently.