Handle null values and missing values correctly in production stats

modelfoxdotdev / modelfox

ModelFox makes it easy to train, deploy, and monitor machine learning models.

Other

1.46k stars 63 forks source link

Handle null values and missing values correctly in production stats #85

Closed isabella closed 2 years ago

isabella commented 2 years ago

Missing values v. null values are handled differently when calling logPrediction Given a model that takes two fields, a and b, the following two json payloads result in different production metrics:

This payload is missing the b field:

{
  a: "hi"
}

This payload has the b field set to null:

{
  a: "hi",
  b: null
}