5.8 Deepnet Limits

Some specific limits apply for your deepnets regarding your dataset field characteristics (see Field Limits ) and the visualization, i.e. to the deepnet Partial Dependence Plot, depending on the number of classes in the objective field and the number of input fields in your dataset (see subsection 5.8.1 ).

Note: the visualization limits just affect to the visualization of the model, i.e., despite your dataset reach those limits, you can still creating the deepnet, evaluating it and using it to make predictions.

Field Limits

Deepnets, similarly to other BigML models, has the following limitations according to the type of field:

• Classes: a maximum number of 1,000 distinct classes per field is allowed.

• Terms: BigML can handle up to 1,000 terms in total. If multiple text fields are defined, then the token limit per field is evenly divided by the number of text fields evenly, e.g., a dataset with two text fields would result in 500 terms per text field. BigML selects those terms with most significant frequency, discarding both those that appear either too often or too infrequently. A maximum of 256 characters per term is allowed.

• Items: a maximum number of 10,000 distinct items per field is allowed.

5.8.1 PDP Limits

There are some circumstances under which your chart cannot be displayed:

• As the PDP only supports numeric and categorical fields for the axes, if your deepnet only contains text, or items fields, the PDP cannot be displayed.

• If your deepnet contains more than 100 fields the top 100 fields will be included as input fields ranked by importance. The rest of fields will be excluded from the view.

• If your deepnet contains more than 200 categories in the objective field, the PDP cannot be displayed.