Classification and Regression with the BigML Dashboard
6.8 Fusion Limits
There are some limits that apply to the creation of any BigML resource. These are limits based on the number of classes, terms and items that can be considered to create your models. In the case of fusions, the single model limits apply along with some specific limits for the fusion creation and visualization:
Maximum models: you can select a maximum of 1,000 models to create your fusion.
PDP limits: the fusion Partial Dependence Plot will not be displayed if:
The fusion only has text or items fields as inputs (because only categorical and numeric fields can be plotted on the PDP axes)
The objective field has more than 200 categories.
The fusion reaches a certain number of models defined by the formula: \((\text{decision trees} + 5\times \text{ensembles} + \text{logistic regressions} + 10\times \text{deepnets}) \leq 50\).
In these cases, the model list will be the default view and the icon to access the PDP will be disabled (see Figure 6.73 ).
Figure 6.73 The PDP cannot be displayed It may happen that the number of component models does not reach the maximum but they are too big to load the fusion predictions in a reasonable time, in this case, BigML will display a warning message (see Figure 6.74 ).
Figure 6.74 The PDP cannot be displayed because the predictions take too much time to be computed Note: the PDP limits only affect the fusion visualization, i.e., even if your dataset reaches those limits, you can still create the fusion, evaluate it and use it to make predictions.
Note2: in the case a fusion has more than 100 input fields, only the top 100 fields ordered by importance will be displayed in the PDP view.