metrics.yml

# `full_name`, `abbr`, and `format` affect metric display in visualizations
# `objective` specifies which direction to go for optimization.

regression:
  ['mse', 'mae', 'r_2']
binary_classification:   # uses macro averaging when available
  ['log_loss', 'accuracy', 'auc', 'precision', 'recall',
  'f1']
classification:   # uses micro averaging when available
  ['log_loss', 'accuracy', 'precision', 'recall',
  'f1',]

# Regression Metrics

mse:
  full_name: Mean Squared Error
  abbr: MSE
  objective: min
mae:
  full_name: Mean Absolute Error
  abbr: MAE
  objective: min
msle:
  full_name: Mean Squared Logarithmic Error
  abbr: MSLE
  objective: min
poisson:
  full_name: Poisson Loss
  abbr: Poisson Loss
  objective: min
r_2:
  full_name: R²
  abbr: R²
  objective: max

# Binary/Classification Metrics

log_loss:
  full_name: Log Loss
  abbr: Log Loss
  objective: min
categorical_crossentropy:
  full_name: Categorical Loss
  abbr: Categorical Loss
  objective: min
binary_crossentropy:
  full_name: Binary Loss
  abbr: Binary Loss
  objective: min
accuracy:
  full_name: Accuracy
  abbr: Accuracy
  objective: max
  format: percent
precision:
  full_name: Precision
  abbr: Precision
  objective: max
  format: percent
recall:
  full_name: Recall
  abbr: Recall
  objective: max
  format: percent
f1:
  full_name: F1 Score
  abbr: F1
  objective: max
  format: percent
auc:
  full_name: Area Under the ROC
  abbr: ROC AUC
  objective: max