Configuring Applications

You can optionally configure your explanation application with a context_config yaml or Python dictionary.

Sample Configuration

A simple context config may look like:

context_config.yaml

# Model prediction output type, one of "numeric", "boolean", "categorical"
output_type: "numeric"
# Python f-string for formatting numeric model outputs
output_format_string: "${:,.2f}"
# If true, increase in model predictions correspond to a positive outcome.
output_sentiment_is_negative: False

# Context-specific overrides for common terminology
terms:
  entity: "House"
  feature: "Factor"
  prediction: "Sale Price"
  positive: "Beneficial"
  negative: "Detrimental"

pages_to_show:
  - Explore a Prediction
  - Similar Entities
  - Experiment with Changes
  - Understand the Model
  - Settings

All Configuration Options

Sibyl-API currently accepts the following configurations (see the context config template for the most current set of configurations).

Explanation option configurations

page_to_show (list): List of explanation pages to show. Sibylapp supports the following list, but other GUIs may offer different options

• Explore a Prediction

• Similar Entities

• Compre Entities

• Experiment with Changes

• Understand the Model

• Settings

allow_page_selection (boolean): Whether of not to allow users to modify the default explanation list

terms (object, string -> string): Dictionary of terms to use. Sibylapp currently supports:

• entity: Inputs to the model

• feature: Information used by the model

• prediction: Output of the model

• positive: Description of features that increase the model's prediction

• negative: Description of features that decrease the model's prediction

use_rows (boolean): Whether to allow users to select rows from entities

row_label (string): How to label individual rows

Output configurations

output_type (one of numeric, boolean, or categorical): Output type of the model

output_pos_label (string): For boolean models, how to refer to positive predictions

output_neg_label (string): For boolean models, how to refer to negative predictions

output_format_string (string): For numeric models, f-string to format outputs

output_sentiment_is_negative (boolean): If True, increases in model predictions correspond to a negative outcome. If False, increases in model predictions correspond to a positive outcome. If None, model outputs are neutral.

show_probs (boolean): For boolean models, whether to show prediction probabilities along with boolean predictions.